x86: early res print out alignment v2
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kernel / e820.c
blobfa77cb4185c3fb12a2e07cce598d4dceed378335
1 /*
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>
19 #include <linux/mm.h>
20 #include <linux/pfn.h>
21 #include <linux/suspend.h>
23 #include <asm/pgtable.h>
24 #include <asm/page.h>
25 #include <asm/e820.h>
26 #include <asm/proto.h>
27 #include <asm/setup.h>
28 #include <asm/trampoline.h>
30 struct e820map e820;
32 /* For PCI or other memory-mapped resources */
33 unsigned long pci_mem_start = 0xaeedbabe;
34 #ifdef CONFIG_PCI
35 EXPORT_SYMBOL(pci_mem_start);
36 #endif
39 * This function checks if any part of the range <start,end> is mapped
40 * with type.
42 int
43 e820_any_mapped(u64 start, u64 end, unsigned type)
45 int i;
47 for (i = 0; i < e820.nr_map; i++) {
48 struct e820entry *ei = &e820.map[i];
50 if (type && ei->type != type)
51 continue;
52 if (ei->addr >= end || ei->addr + ei->size <= start)
53 continue;
54 return 1;
56 return 0;
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)
68 int i;
70 for (i = 0; i < e820.nr_map; i++) {
71 struct e820entry *ei = &e820.map[i];
73 if (type && ei->type != type)
74 continue;
75 /* is the region (part) in overlap with the current region ?*/
76 if (ei->addr >= end || ei->addr + ei->size <= start)
77 continue;
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
86 * coverage
88 if (start >= end)
89 return 1;
91 return 0;
95 * Add a memory region to the kernel e820 map.
97 void __init e820_add_region(u64 start, u64 size, int type)
99 int x = e820.nr_map;
101 if (x == ARRAY_SIZE(e820.map)) {
102 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
103 return;
106 e820.map[x].addr = start;
107 e820.map[x].size = size;
108 e820.map[x].type = type;
109 e820.nr_map++;
112 void __init e820_print_map(char *who)
114 int i;
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,
119 (unsigned long long)
120 (e820.map[i].addr + e820.map[i].size));
121 switch (e820.map[i].type) {
122 case E820_RAM:
123 printk(KERN_CONT "(usable)\n");
124 break;
125 case E820_RESERVED:
126 printk(KERN_CONT "(reserved)\n");
127 break;
128 case E820_ACPI:
129 printk(KERN_CONT "(ACPI data)\n");
130 break;
131 case E820_NVS:
132 printk(KERN_CONT "(ACPI NVS)\n");
133 break;
134 default:
135 printk(KERN_CONT "type %u\n", e820.map[i].type);
136 break;
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
147 * numbered type.
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,
204 int *pnr_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;
218 int overlap_entries;
219 int new_bios_entry;
220 int old_nr, new_nr, chg_nr;
221 int i;
223 /* if there's only one memory region, don't bother */
224 if (*pnr_map < 2)
225 return -1;
227 old_nr = *pnr_map;
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)
233 return -1;
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 */
241 chgidx = 0;
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 +
247 biosmap[i].size;
248 change_point[chgidx++]->pbios = &biosmap[i];
251 chg_nr = chgidx;
253 /* sort change-point list by memory addresses (low -> high) */
254 still_changing = 1;
255 while (still_changing) {
256 still_changing = 0;
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;
279 still_changing = 1;
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;
301 } else {
303 * remove entry from list (order independent,
304 * so swap with last)
306 for (i = 0; i < overlap_entries; i++) {
307 if (overlap_list[i] ==
308 change_point[chgidx]->pbios)
309 overlap_list[i] =
310 overlap_list[overlap_entries-1];
312 overlap_entries--;
315 * if there are overlapping entries, decide which
316 * "type" to use (larger value takes precedence --
317 * 1=usable, 2,3,4,4+=unusable)
319 current_type = 0;
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
325 * information
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
333 * was non-zero
335 if (new_bios[new_bios_entry].size != 0)
337 * no more space left for new
338 * bios entries ?
340 if (++new_bios_entry >= max_nr_map)
341 break;
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));
357 *pnr_map = new_nr;
359 return 0;
362 static int __init __copy_e820_map(struct e820entry *biosmap, int nr_map)
364 while (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. */
371 if (start > end)
372 return -1;
374 e820_add_region(start, size, type);
376 biosmap++;
377 nr_map--;
379 return 0;
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 */
394 if (nr_map < 2)
395 return -1;
397 return __copy_e820_map(biosmap, nr_map);
400 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
401 unsigned new_type)
403 int i;
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)
415 continue;
416 /* totally covered? */
417 if (ei->addr >= start &&
418 (ei->addr + ei->size) <= (start + size)) {
419 ei->type = new_type;
420 real_updated_size += ei->size;
421 continue;
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)
427 continue;
428 e820_add_region(final_start, final_end - final_start,
429 new_type);
430 real_updated_size += final_end - final_start;
432 ei->size -= final_end - final_start;
433 if (ei->addr < final_start)
434 continue;
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,
442 int checktype)
444 int i;
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)
455 continue;
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));
461 continue;
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)
467 continue;
468 real_removed_size += final_end - final_start;
470 ei->size -= final_end - final_start;
471 if (ei->addr < final_start)
472 continue;
473 ei->addr = final_end;
475 return real_removed_size;
478 void __init update_e820(void)
480 int nr_map;
482 nr_map = e820.nr_map;
483 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
484 return;
485 e820.nr_map = nr_map;
486 printk(KERN_INFO "modified physical RAM map:\n");
487 e820_print_map("modified");
491 * Search for a gap in the e820 memory space from start_addr to 2^32.
493 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
494 unsigned long start_addr)
496 unsigned long long last = 0x100000000ull;
497 int i = e820.nr_map;
498 int found = 0;
500 while (--i >= 0) {
501 unsigned long long start = e820.map[i].addr;
502 unsigned long long end = start + e820.map[i].size;
504 if (end < start_addr)
505 continue;
508 * Since "last" is at most 4GB, we know we'll
509 * fit in 32 bits if this condition is true
511 if (last > end) {
512 unsigned long gap = last - end;
514 if (gap >= *gapsize) {
515 *gapsize = gap;
516 *gapstart = end;
517 found = 1;
520 if (start < last)
521 last = start;
523 return found;
527 * Search for the biggest gap in the low 32 bits of the e820
528 * memory space. We pass this space to PCI to assign MMIO resources
529 * for hotplug or unconfigured devices in.
530 * Hopefully the BIOS let enough space left.
532 __init void e820_setup_gap(void)
534 unsigned long gapstart, gapsize, round;
535 int found;
537 gapstart = 0x10000000;
538 gapsize = 0x400000;
539 found = e820_search_gap(&gapstart, &gapsize, 0);
541 #ifdef CONFIG_X86_64
542 if (!found) {
543 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
544 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
545 "address range\n"
546 KERN_ERR "PCI: Unassigned devices with 32bit resource "
547 "registers may break!\n");
549 #endif
552 * See how much we want to round up: start off with
553 * rounding to the next 1MB area.
555 round = 0x100000;
556 while ((gapsize >> 4) > round)
557 round += round;
558 /* Fun with two's complement */
559 pci_mem_start = (gapstart + round) & -round;
561 printk(KERN_INFO
562 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
563 pci_mem_start, gapstart, gapsize);
567 * Because of the size limitation of struct boot_params, only first
568 * 128 E820 memory entries are passed to kernel via
569 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
570 * linked list of struct setup_data, which is parsed here.
572 void __init parse_e820_ext(struct setup_data *sdata, unsigned long pa_data)
574 u32 map_len;
575 int entries;
576 struct e820entry *extmap;
578 entries = sdata->len / sizeof(struct e820entry);
579 map_len = sdata->len + sizeof(struct setup_data);
580 if (map_len > PAGE_SIZE)
581 sdata = early_ioremap(pa_data, map_len);
582 extmap = (struct e820entry *)(sdata->data);
583 __copy_e820_map(extmap, entries);
584 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
585 if (map_len > PAGE_SIZE)
586 early_iounmap(sdata, map_len);
587 printk(KERN_INFO "extended physical RAM map:\n");
588 e820_print_map("extended");
591 #if defined(CONFIG_X86_64) || \
592 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
594 * Find the ranges of physical addresses that do not correspond to
595 * e820 RAM areas and mark the corresponding pages as nosave for
596 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
598 * This function requires the e820 map to be sorted and without any
599 * overlapping entries and assumes the first e820 area to be RAM.
601 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
603 int i;
604 unsigned long pfn;
606 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
607 for (i = 1; i < e820.nr_map; i++) {
608 struct e820entry *ei = &e820.map[i];
610 if (pfn < PFN_UP(ei->addr))
611 register_nosave_region(pfn, PFN_UP(ei->addr));
613 pfn = PFN_DOWN(ei->addr + ei->size);
614 if (ei->type != E820_RAM)
615 register_nosave_region(PFN_UP(ei->addr), pfn);
617 if (pfn >= limit_pfn)
618 break;
621 #endif
624 * Early reserved memory areas.
626 #define MAX_EARLY_RES 20
628 struct early_res {
629 u64 start, end;
630 char name[16];
631 char overlap_ok;
633 static struct early_res early_res[MAX_EARLY_RES] __initdata = {
634 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
635 #if defined(CONFIG_X86_64) && defined(CONFIG_X86_TRAMPOLINE)
636 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + 2 * PAGE_SIZE, "TRAMPOLINE" },
637 #endif
638 #if defined(CONFIG_X86_32) && defined(CONFIG_SMP)
640 * But first pinch a few for the stack/trampoline stuff
641 * FIXME: Don't need the extra page at 4K, but need to fix
642 * trampoline before removing it. (see the GDT stuff)
644 { PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE" },
646 * Has to be in very low memory so we can execute
647 * real-mode AP code.
649 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + PAGE_SIZE, "TRAMPOLINE" },
650 #endif
654 static int __init find_overlapped_early(u64 start, u64 end)
656 int i;
657 struct early_res *r;
659 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
660 r = &early_res[i];
661 if (end > r->start && start < r->end)
662 break;
665 return i;
669 * Drop the i-th range from the early reservation map,
670 * by copying any higher ranges down one over it, and
671 * clearing what had been the last slot.
673 static void __init drop_range(int i)
675 int j;
677 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
680 memmove(&early_res[i], &early_res[i + 1],
681 (j - 1 - i) * sizeof(struct early_res));
683 early_res[j - 1].end = 0;
687 * Split any existing ranges that:
688 * 1) are marked 'overlap_ok', and
689 * 2) overlap with the stated range [start, end)
690 * into whatever portion (if any) of the existing range is entirely
691 * below or entirely above the stated range. Drop the portion
692 * of the existing range that overlaps with the stated range,
693 * which will allow the caller of this routine to then add that
694 * stated range without conflicting with any existing range.
696 static void __init drop_overlaps_that_are_ok(u64 start, u64 end)
698 int i;
699 struct early_res *r;
700 u64 lower_start, lower_end;
701 u64 upper_start, upper_end;
702 char name[16];
704 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
705 r = &early_res[i];
707 /* Continue past non-overlapping ranges */
708 if (end <= r->start || start >= r->end)
709 continue;
712 * Leave non-ok overlaps as is; let caller
713 * panic "Overlapping early reservations"
714 * when it hits this overlap.
716 if (!r->overlap_ok)
717 return;
720 * We have an ok overlap. We will drop it from the early
721 * reservation map, and add back in any non-overlapping
722 * portions (lower or upper) as separate, overlap_ok,
723 * non-overlapping ranges.
726 /* 1. Note any non-overlapping (lower or upper) ranges. */
727 strncpy(name, r->name, sizeof(name) - 1);
729 lower_start = lower_end = 0;
730 upper_start = upper_end = 0;
731 if (r->start < start) {
732 lower_start = r->start;
733 lower_end = start;
735 if (r->end > end) {
736 upper_start = end;
737 upper_end = r->end;
740 /* 2. Drop the original ok overlapping range */
741 drop_range(i);
743 i--; /* resume for-loop on copied down entry */
745 /* 3. Add back in any non-overlapping ranges. */
746 if (lower_end)
747 reserve_early_overlap_ok(lower_start, lower_end, name);
748 if (upper_end)
749 reserve_early_overlap_ok(upper_start, upper_end, name);
753 static void __init __reserve_early(u64 start, u64 end, char *name,
754 int overlap_ok)
756 int i;
757 struct early_res *r;
759 i = find_overlapped_early(start, end);
760 if (i >= MAX_EARLY_RES)
761 panic("Too many early reservations");
762 r = &early_res[i];
763 if (r->end)
764 panic("Overlapping early reservations "
765 "%llx-%llx %s to %llx-%llx %s\n",
766 start, end - 1, name?name:"", r->start,
767 r->end - 1, r->name);
768 r->start = start;
769 r->end = end;
770 r->overlap_ok = overlap_ok;
771 if (name)
772 strncpy(r->name, name, sizeof(r->name) - 1);
776 * A few early reservtations come here.
778 * The 'overlap_ok' in the name of this routine does -not- mean it
779 * is ok for these reservations to overlap an earlier reservation.
780 * Rather it means that it is ok for subsequent reservations to
781 * overlap this one.
783 * Use this entry point to reserve early ranges when you are doing
784 * so out of "Paranoia", reserving perhaps more memory than you need,
785 * just in case, and don't mind a subsequent overlapping reservation
786 * that is known to be needed.
788 * The drop_overlaps_that_are_ok() call here isn't really needed.
789 * It would be needed if we had two colliding 'overlap_ok'
790 * reservations, so that the second such would not panic on the
791 * overlap with the first. We don't have any such as of this
792 * writing, but might as well tolerate such if it happens in
793 * the future.
795 void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
797 drop_overlaps_that_are_ok(start, end);
798 __reserve_early(start, end, name, 1);
802 * Most early reservations come here.
804 * We first have drop_overlaps_that_are_ok() drop any pre-existing
805 * 'overlap_ok' ranges, so that we can then reserve this memory
806 * range without risk of panic'ing on an overlapping overlap_ok
807 * early reservation.
809 void __init reserve_early(u64 start, u64 end, char *name)
811 drop_overlaps_that_are_ok(start, end);
812 __reserve_early(start, end, name, 0);
815 void __init free_early(u64 start, u64 end)
817 struct early_res *r;
818 int i;
820 i = find_overlapped_early(start, end);
821 r = &early_res[i];
822 if (i >= MAX_EARLY_RES || r->end != end || r->start != start)
823 panic("free_early on not reserved area: %llx-%llx!",
824 start, end - 1);
826 drop_range(i);
829 void __init early_res_to_bootmem(u64 start, u64 end)
831 int i, count;
832 u64 final_start, final_end;
834 count = 0;
835 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++)
836 count++;
838 printk(KERN_INFO "(%d early reservations) ==> bootmem\n", count);
839 for (i = 0; i < count; i++) {
840 struct early_res *r = &early_res[i];
841 printk(KERN_INFO " #%d [ %010llx - %010llx ] %16s", i,
842 r->start, r->end, r->name);
843 final_start = max(start, r->start);
844 final_end = min(end, r->end);
845 if (final_start >= final_end) {
846 printk(KERN_CONT "\n");
847 continue;
849 printk(KERN_CONT " ===> [ %010llx - %010llx ]\n",
850 final_start, final_end);
851 reserve_bootmem_generic(final_start, final_end - final_start,
852 BOOTMEM_DEFAULT);
856 /* Check for already reserved areas */
857 static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
859 int i;
860 u64 addr = *addrp;
861 int changed = 0;
862 struct early_res *r;
863 again:
864 i = find_overlapped_early(addr, addr + size);
865 r = &early_res[i];
866 if (i < MAX_EARLY_RES && r->end) {
867 *addrp = addr = round_up(r->end, align);
868 changed = 1;
869 goto again;
871 return changed;
874 /* Check for already reserved areas */
875 static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
877 int i;
878 u64 addr = *addrp, last;
879 u64 size = *sizep;
880 int changed = 0;
881 again:
882 last = addr + size;
883 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
884 struct early_res *r = &early_res[i];
885 if (last > r->start && addr < r->start) {
886 size = r->start - addr;
887 changed = 1;
888 goto again;
890 if (last > r->end && addr < r->end) {
891 addr = round_up(r->end, align);
892 size = last - addr;
893 changed = 1;
894 goto again;
896 if (last <= r->end && addr >= r->start) {
897 (*sizep)++;
898 return 0;
901 if (changed) {
902 *addrp = addr;
903 *sizep = size;
905 return changed;
909 * Find a free area with specified alignment in a specific range.
911 u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
913 int i;
915 for (i = 0; i < e820.nr_map; i++) {
916 struct e820entry *ei = &e820.map[i];
917 u64 addr, last;
918 u64 ei_last;
920 if (ei->type != E820_RAM)
921 continue;
922 addr = round_up(ei->addr, align);
923 ei_last = ei->addr + ei->size;
924 if (addr < start)
925 addr = round_up(start, align);
926 if (addr >= ei_last)
927 continue;
928 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
930 last = addr + size;
931 if (last > ei_last)
932 continue;
933 if (last > end)
934 continue;
935 return addr;
937 return -1ULL;
941 * Find next free range after *start
943 u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
945 int i;
947 for (i = 0; i < e820.nr_map; i++) {
948 struct e820entry *ei = &e820.map[i];
949 u64 addr, last;
950 u64 ei_last;
952 if (ei->type != E820_RAM)
953 continue;
954 addr = round_up(ei->addr, align);
955 ei_last = ei->addr + ei->size;
956 if (addr < start)
957 addr = round_up(start, align);
958 if (addr >= ei_last)
959 continue;
960 *sizep = ei_last - addr;
961 while (bad_addr_size(&addr, sizep, align) &&
962 addr + *sizep <= ei_last)
964 last = addr + *sizep;
965 if (last > ei_last)
966 continue;
967 return addr;
969 return -1UL;
974 * pre allocated 4k and reserved it in e820
976 u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
978 u64 size = 0;
979 u64 addr;
980 u64 start;
982 start = startt;
983 while (size < sizet)
984 start = find_e820_area_size(start, &size, align);
986 if (size < sizet)
987 return 0;
989 addr = round_down(start + size - sizet, align);
990 e820_update_range(addr, sizet, E820_RAM, E820_RESERVED);
991 printk(KERN_INFO "update e820 for early_reserve_e820\n");
992 update_e820();
994 return addr;
997 #ifdef CONFIG_X86_32
998 # ifdef CONFIG_X86_PAE
999 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
1000 # else
1001 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
1002 # endif
1003 #else /* CONFIG_X86_32 */
1004 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
1005 #endif
1008 * Last pfn which the user wants to use.
1010 unsigned long __initdata end_user_pfn = MAX_ARCH_PFN;
1013 * Find the highest page frame number we have available
1015 unsigned long __init e820_end_of_ram(void)
1017 unsigned long last_pfn;
1018 unsigned long max_arch_pfn = MAX_ARCH_PFN;
1020 last_pfn = find_max_pfn_with_active_regions();
1022 if (last_pfn > max_arch_pfn)
1023 last_pfn = max_arch_pfn;
1024 if (last_pfn > end_user_pfn)
1025 last_pfn = end_user_pfn;
1027 printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
1028 last_pfn, max_arch_pfn);
1029 return last_pfn;
1033 * Finds an active region in the address range from start_pfn to last_pfn and
1034 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
1036 int __init e820_find_active_region(const struct e820entry *ei,
1037 unsigned long start_pfn,
1038 unsigned long last_pfn,
1039 unsigned long *ei_startpfn,
1040 unsigned long *ei_endpfn)
1042 u64 align = PAGE_SIZE;
1044 *ei_startpfn = round_up(ei->addr, align) >> PAGE_SHIFT;
1045 *ei_endpfn = round_down(ei->addr + ei->size, align) >> PAGE_SHIFT;
1047 /* Skip map entries smaller than a page */
1048 if (*ei_startpfn >= *ei_endpfn)
1049 return 0;
1051 /* Skip if map is outside the node */
1052 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
1053 *ei_startpfn >= last_pfn)
1054 return 0;
1056 /* Check for overlaps */
1057 if (*ei_startpfn < start_pfn)
1058 *ei_startpfn = start_pfn;
1059 if (*ei_endpfn > last_pfn)
1060 *ei_endpfn = last_pfn;
1062 /* Obey end_user_pfn to save on memmap */
1063 if (*ei_startpfn >= end_user_pfn)
1064 return 0;
1065 if (*ei_endpfn > end_user_pfn)
1066 *ei_endpfn = end_user_pfn;
1068 return 1;
1071 /* Walk the e820 map and register active regions within a node */
1072 void __init e820_register_active_regions(int nid, unsigned long start_pfn,
1073 unsigned long last_pfn)
1075 unsigned long ei_startpfn;
1076 unsigned long ei_endpfn;
1077 int i;
1079 for (i = 0; i < e820.nr_map; i++)
1080 if (e820_find_active_region(&e820.map[i],
1081 start_pfn, last_pfn,
1082 &ei_startpfn, &ei_endpfn))
1083 add_active_range(nid, ei_startpfn, ei_endpfn);
1087 * Find the hole size (in bytes) in the memory range.
1088 * @start: starting address of the memory range to scan
1089 * @end: ending address of the memory range to scan
1091 u64 __init e820_hole_size(u64 start, u64 end)
1093 unsigned long start_pfn = start >> PAGE_SHIFT;
1094 unsigned long last_pfn = end >> PAGE_SHIFT;
1095 unsigned long ei_startpfn, ei_endpfn, ram = 0;
1096 int i;
1098 for (i = 0; i < e820.nr_map; i++) {
1099 if (e820_find_active_region(&e820.map[i],
1100 start_pfn, last_pfn,
1101 &ei_startpfn, &ei_endpfn))
1102 ram += ei_endpfn - ei_startpfn;
1104 return end - start - ((u64)ram << PAGE_SHIFT);
1107 static void early_panic(char *msg)
1109 early_printk(msg);
1110 panic(msg);
1113 /* "mem=nopentium" disables the 4MB page tables. */
1114 static int __init parse_memopt(char *p)
1116 u64 mem_size;
1118 if (!p)
1119 return -EINVAL;
1121 #ifdef CONFIG_X86_32
1122 if (!strcmp(p, "nopentium")) {
1123 setup_clear_cpu_cap(X86_FEATURE_PSE);
1124 return 0;
1126 #endif
1128 mem_size = memparse(p, &p);
1129 end_user_pfn = mem_size>>PAGE_SHIFT;
1130 e820_update_range(mem_size, ULLONG_MAX - mem_size,
1131 E820_RAM, E820_RESERVED);
1133 return 0;
1135 early_param("mem", parse_memopt);
1137 static int userdef __initdata;
1139 static int __init parse_memmap_opt(char *p)
1141 char *oldp;
1142 u64 start_at, mem_size;
1144 if (!strcmp(p, "exactmap")) {
1145 #ifdef CONFIG_CRASH_DUMP
1147 * If we are doing a crash dump, we still need to know
1148 * the real mem size before original memory map is
1149 * reset.
1151 e820_register_active_regions(0, 0, -1UL);
1152 saved_max_pfn = e820_end_of_ram();
1153 remove_all_active_ranges();
1154 #endif
1155 e820.nr_map = 0;
1156 userdef = 1;
1157 return 0;
1160 oldp = p;
1161 mem_size = memparse(p, &p);
1162 if (p == oldp)
1163 return -EINVAL;
1165 userdef = 1;
1166 if (*p == '@') {
1167 start_at = memparse(p+1, &p);
1168 e820_add_region(start_at, mem_size, E820_RAM);
1169 } else if (*p == '#') {
1170 start_at = memparse(p+1, &p);
1171 e820_add_region(start_at, mem_size, E820_ACPI);
1172 } else if (*p == '$') {
1173 start_at = memparse(p+1, &p);
1174 e820_add_region(start_at, mem_size, E820_RESERVED);
1175 } else {
1176 end_user_pfn = (mem_size >> PAGE_SHIFT);
1177 e820_update_range(mem_size, ULLONG_MAX - mem_size,
1178 E820_RAM, E820_RESERVED);
1180 return *p == '\0' ? 0 : -EINVAL;
1182 early_param("memmap", parse_memmap_opt);
1184 void __init finish_e820_parsing(void)
1186 if (userdef) {
1187 int nr = e820.nr_map;
1189 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
1190 early_panic("Invalid user supplied memory map");
1191 e820.nr_map = nr;
1193 printk(KERN_INFO "user-defined physical RAM map:\n");
1194 e820_print_map("user");
1199 * Mark e820 reserved areas as busy for the resource manager.
1201 void __init e820_reserve_resources(void)
1203 int i;
1204 struct resource *res;
1206 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
1207 for (i = 0; i < e820.nr_map; i++) {
1208 switch (e820.map[i].type) {
1209 case E820_RAM: res->name = "System RAM"; break;
1210 case E820_ACPI: res->name = "ACPI Tables"; break;
1211 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
1212 default: res->name = "reserved";
1214 res->start = e820.map[i].addr;
1215 res->end = res->start + e820.map[i].size - 1;
1216 #ifndef CONFIG_RESOURCES_64BIT
1217 if (res->end > 0x100000000ULL) {
1218 res++;
1219 continue;
1221 #endif
1222 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1223 insert_resource(&iomem_resource, res);
1224 res++;
1228 char *__init default_machine_specific_memory_setup(void)
1230 char *who = "BIOS-e820";
1231 int new_nr;
1233 * Try to copy the BIOS-supplied E820-map.
1235 * Otherwise fake a memory map; one section from 0k->640k,
1236 * the next section from 1mb->appropriate_mem_k
1238 new_nr = boot_params.e820_entries;
1239 sanitize_e820_map(boot_params.e820_map,
1240 ARRAY_SIZE(boot_params.e820_map),
1241 &new_nr);
1242 boot_params.e820_entries = new_nr;
1243 if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0) {
1244 u64 mem_size;
1246 /* compare results from other methods and take the greater */
1247 if (boot_params.alt_mem_k
1248 < boot_params.screen_info.ext_mem_k) {
1249 mem_size = boot_params.screen_info.ext_mem_k;
1250 who = "BIOS-88";
1251 } else {
1252 mem_size = boot_params.alt_mem_k;
1253 who = "BIOS-e801";
1256 e820.nr_map = 0;
1257 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1258 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1261 /* In case someone cares... */
1262 return who;
1265 char *__init __attribute__((weak)) machine_specific_memory_setup(void)
1267 return default_machine_specific_memory_setup();
1270 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
1271 char * __init __attribute__((weak)) memory_setup(void)
1273 return machine_specific_memory_setup();
1276 void __init setup_memory_map(void)
1278 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1279 e820_print_map(memory_setup());
1282 #ifdef CONFIG_X86_64
1283 int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
1285 int i;
1287 if (slot < 0 || slot >= e820.nr_map)
1288 return -1;
1289 for (i = slot; i < e820.nr_map; i++) {
1290 if (e820.map[i].type != E820_RAM)
1291 continue;
1292 break;
1294 if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
1295 return -1;
1296 *addr = e820.map[i].addr;
1297 *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
1298 max_pfn << PAGE_SHIFT) - *addr;
1299 return i + 1;
1301 #endif