checkpatch: dissallow spaces between stars in pointer types
[linux-2.6/mini2440.git] / arch / x86 / kernel / e820.c
blob65a13943e0982524877768164cf8cc0cd63688f4
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>
22 #include <linux/firmware-map.h>
24 #include <asm/pgtable.h>
25 #include <asm/page.h>
26 #include <asm/e820.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.
45 struct e820map e820;
46 struct e820map e820_saved;
48 /* For PCI or other memory-mapped resources */
49 unsigned long pci_mem_start = 0xaeedbabe;
50 #ifdef CONFIG_PCI
51 EXPORT_SYMBOL(pci_mem_start);
52 #endif
55 * This function checks if any part of the range <start,end> is mapped
56 * with type.
58 int
59 e820_any_mapped(u64 start, u64 end, unsigned type)
61 int i;
63 for (i = 0; i < e820.nr_map; i++) {
64 struct e820entry *ei = &e820.map[i];
66 if (type && ei->type != type)
67 continue;
68 if (ei->addr >= end || ei->addr + ei->size <= start)
69 continue;
70 return 1;
72 return 0;
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 __init e820_all_mapped(u64 start, u64 end, unsigned type)
84 int i;
86 for (i = 0; i < e820.nr_map; i++) {
87 struct e820entry *ei = &e820.map[i];
89 if (type && ei->type != type)
90 continue;
91 /* is the region (part) in overlap with the current region ?*/
92 if (ei->addr >= end || ei->addr + ei->size <= start)
93 continue;
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
102 * coverage
104 if (start >= end)
105 return 1;
107 return 0;
111 * Add a memory region to the kernel e820 map.
113 void __init e820_add_region(u64 start, u64 size, int type)
115 int x = e820.nr_map;
117 if (x == ARRAY_SIZE(e820.map)) {
118 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
119 return;
122 e820.map[x].addr = start;
123 e820.map[x].size = size;
124 e820.map[x].type = type;
125 e820.nr_map++;
128 void __init e820_print_map(char *who)
130 int i;
132 for (i = 0; i < e820.nr_map; i++) {
133 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
134 (unsigned long long) e820.map[i].addr,
135 (unsigned long long)
136 (e820.map[i].addr + e820.map[i].size));
137 switch (e820.map[i].type) {
138 case E820_RAM:
139 case E820_RESERVED_KERN:
140 printk(KERN_CONT "(usable)\n");
141 break;
142 case E820_RESERVED:
143 printk(KERN_CONT "(reserved)\n");
144 break;
145 case E820_ACPI:
146 printk(KERN_CONT "(ACPI data)\n");
147 break;
148 case E820_NVS:
149 printk(KERN_CONT "(ACPI NVS)\n");
150 break;
151 case E820_UNUSABLE:
152 printk("(unusable)\n");
153 break;
154 default:
155 printk(KERN_CONT "type %u\n", e820.map[i].type);
156 break;
162 * Sanitize the BIOS e820 map.
164 * Some e820 responses include overlapping entries. The following
165 * replaces the original e820 map with a new one, removing overlaps,
166 * and resolving conflicting memory types in favor of highest
167 * numbered type.
169 * The input parameter biosmap points to an array of 'struct
170 * e820entry' which on entry has elements in the range [0, *pnr_map)
171 * valid, and which has space for up to max_nr_map entries.
172 * On return, the resulting sanitized e820 map entries will be in
173 * overwritten in the same location, starting at biosmap.
175 * The integer pointed to by pnr_map must be valid on entry (the
176 * current number of valid entries located at biosmap) and will
177 * be updated on return, with the new number of valid entries
178 * (something no more than max_nr_map.)
180 * The return value from sanitize_e820_map() is zero if it
181 * successfully 'sanitized' the map entries passed in, and is -1
182 * if it did nothing, which can happen if either of (1) it was
183 * only passed one map entry, or (2) any of the input map entries
184 * were invalid (start + size < start, meaning that the size was
185 * so big the described memory range wrapped around through zero.)
187 * Visually we're performing the following
188 * (1,2,3,4 = memory types)...
190 * Sample memory map (w/overlaps):
191 * ____22__________________
192 * ______________________4_
193 * ____1111________________
194 * _44_____________________
195 * 11111111________________
196 * ____________________33__
197 * ___________44___________
198 * __________33333_________
199 * ______________22________
200 * ___________________2222_
201 * _________111111111______
202 * _____________________11_
203 * _________________4______
205 * Sanitized equivalent (no overlap):
206 * 1_______________________
207 * _44_____________________
208 * ___1____________________
209 * ____22__________________
210 * ______11________________
211 * _________1______________
212 * __________3_____________
213 * ___________44___________
214 * _____________33_________
215 * _______________2________
216 * ________________1_______
217 * _________________4______
218 * ___________________2____
219 * ____________________33__
220 * ______________________4_
223 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
224 int *pnr_map)
226 struct change_member {
227 struct e820entry *pbios; /* pointer to original bios entry */
228 unsigned long long addr; /* address for this change point */
230 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
231 static struct change_member *change_point[2*E820_X_MAX] __initdata;
232 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
233 static struct e820entry new_bios[E820_X_MAX] __initdata;
234 struct change_member *change_tmp;
235 unsigned long current_type, last_type;
236 unsigned long long last_addr;
237 int chgidx, still_changing;
238 int overlap_entries;
239 int new_bios_entry;
240 int old_nr, new_nr, chg_nr;
241 int i;
243 /* if there's only one memory region, don't bother */
244 if (*pnr_map < 2)
245 return -1;
247 old_nr = *pnr_map;
248 BUG_ON(old_nr > max_nr_map);
250 /* bail out if we find any unreasonable addresses in bios map */
251 for (i = 0; i < old_nr; i++)
252 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
253 return -1;
255 /* create pointers for initial change-point information (for sorting) */
256 for (i = 0; i < 2 * old_nr; i++)
257 change_point[i] = &change_point_list[i];
259 /* record all known change-points (starting and ending addresses),
260 omitting those that are for empty memory regions */
261 chgidx = 0;
262 for (i = 0; i < old_nr; i++) {
263 if (biosmap[i].size != 0) {
264 change_point[chgidx]->addr = biosmap[i].addr;
265 change_point[chgidx++]->pbios = &biosmap[i];
266 change_point[chgidx]->addr = biosmap[i].addr +
267 biosmap[i].size;
268 change_point[chgidx++]->pbios = &biosmap[i];
271 chg_nr = chgidx;
273 /* sort change-point list by memory addresses (low -> high) */
274 still_changing = 1;
275 while (still_changing) {
276 still_changing = 0;
277 for (i = 1; i < chg_nr; i++) {
278 unsigned long long curaddr, lastaddr;
279 unsigned long long curpbaddr, lastpbaddr;
281 curaddr = change_point[i]->addr;
282 lastaddr = change_point[i - 1]->addr;
283 curpbaddr = change_point[i]->pbios->addr;
284 lastpbaddr = change_point[i - 1]->pbios->addr;
287 * swap entries, when:
289 * curaddr > lastaddr or
290 * curaddr == lastaddr and curaddr == curpbaddr and
291 * lastaddr != lastpbaddr
293 if (curaddr < lastaddr ||
294 (curaddr == lastaddr && curaddr == curpbaddr &&
295 lastaddr != lastpbaddr)) {
296 change_tmp = change_point[i];
297 change_point[i] = change_point[i-1];
298 change_point[i-1] = change_tmp;
299 still_changing = 1;
304 /* create a new bios memory map, removing overlaps */
305 overlap_entries = 0; /* number of entries in the overlap table */
306 new_bios_entry = 0; /* index for creating new bios map entries */
307 last_type = 0; /* start with undefined memory type */
308 last_addr = 0; /* start with 0 as last starting address */
310 /* loop through change-points, determining affect on the new bios map */
311 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
312 /* keep track of all overlapping bios entries */
313 if (change_point[chgidx]->addr ==
314 change_point[chgidx]->pbios->addr) {
316 * add map entry to overlap list (> 1 entry
317 * implies an overlap)
319 overlap_list[overlap_entries++] =
320 change_point[chgidx]->pbios;
321 } else {
323 * remove entry from list (order independent,
324 * so swap with last)
326 for (i = 0; i < overlap_entries; i++) {
327 if (overlap_list[i] ==
328 change_point[chgidx]->pbios)
329 overlap_list[i] =
330 overlap_list[overlap_entries-1];
332 overlap_entries--;
335 * if there are overlapping entries, decide which
336 * "type" to use (larger value takes precedence --
337 * 1=usable, 2,3,4,4+=unusable)
339 current_type = 0;
340 for (i = 0; i < overlap_entries; i++)
341 if (overlap_list[i]->type > current_type)
342 current_type = overlap_list[i]->type;
344 * continue building up new bios map based on this
345 * information
347 if (current_type != last_type) {
348 if (last_type != 0) {
349 new_bios[new_bios_entry].size =
350 change_point[chgidx]->addr - last_addr;
352 * move forward only if the new size
353 * was non-zero
355 if (new_bios[new_bios_entry].size != 0)
357 * no more space left for new
358 * bios entries ?
360 if (++new_bios_entry >= max_nr_map)
361 break;
363 if (current_type != 0) {
364 new_bios[new_bios_entry].addr =
365 change_point[chgidx]->addr;
366 new_bios[new_bios_entry].type = current_type;
367 last_addr = change_point[chgidx]->addr;
369 last_type = current_type;
372 /* retain count for new bios entries */
373 new_nr = new_bios_entry;
375 /* copy new bios mapping into original location */
376 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
377 *pnr_map = new_nr;
379 return 0;
382 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
384 while (nr_map) {
385 u64 start = biosmap->addr;
386 u64 size = biosmap->size;
387 u64 end = start + size;
388 u32 type = biosmap->type;
390 /* Overflow in 64 bits? Ignore the memory map. */
391 if (start > end)
392 return -1;
394 e820_add_region(start, size, type);
396 biosmap++;
397 nr_map--;
399 return 0;
403 * Copy the BIOS e820 map into a safe place.
405 * Sanity-check it while we're at it..
407 * If we're lucky and live on a modern system, the setup code
408 * will have given us a memory map that we can use to properly
409 * set up memory. If we aren't, we'll fake a memory map.
411 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
413 /* Only one memory region (or negative)? Ignore it */
414 if (nr_map < 2)
415 return -1;
417 return __append_e820_map(biosmap, nr_map);
420 static u64 __init e820_update_range_map(struct e820map *e820x, u64 start,
421 u64 size, unsigned old_type,
422 unsigned new_type)
424 int i;
425 u64 real_updated_size = 0;
427 BUG_ON(old_type == new_type);
429 if (size > (ULLONG_MAX - start))
430 size = ULLONG_MAX - start;
432 for (i = 0; i < e820.nr_map; i++) {
433 struct e820entry *ei = &e820x->map[i];
434 u64 final_start, final_end;
435 if (ei->type != old_type)
436 continue;
437 /* totally covered? */
438 if (ei->addr >= start &&
439 (ei->addr + ei->size) <= (start + size)) {
440 ei->type = new_type;
441 real_updated_size += ei->size;
442 continue;
444 /* partially covered */
445 final_start = max(start, ei->addr);
446 final_end = min(start + size, ei->addr + ei->size);
447 if (final_start >= final_end)
448 continue;
449 e820_add_region(final_start, final_end - final_start,
450 new_type);
451 real_updated_size += final_end - final_start;
453 ei->size -= final_end - final_start;
454 if (ei->addr < final_start)
455 continue;
456 ei->addr = final_end;
458 return real_updated_size;
461 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
462 unsigned new_type)
464 return e820_update_range_map(&e820, start, size, old_type, new_type);
467 static u64 __init e820_update_range_saved(u64 start, u64 size,
468 unsigned old_type, unsigned new_type)
470 return e820_update_range_map(&e820_saved, start, size, old_type,
471 new_type);
474 /* make e820 not cover the range */
475 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
476 int checktype)
478 int i;
479 u64 real_removed_size = 0;
481 if (size > (ULLONG_MAX - start))
482 size = ULLONG_MAX - start;
484 for (i = 0; i < e820.nr_map; i++) {
485 struct e820entry *ei = &e820.map[i];
486 u64 final_start, final_end;
488 if (checktype && ei->type != old_type)
489 continue;
490 /* totally covered? */
491 if (ei->addr >= start &&
492 (ei->addr + ei->size) <= (start + size)) {
493 real_removed_size += ei->size;
494 memset(ei, 0, sizeof(struct e820entry));
495 continue;
497 /* partially covered */
498 final_start = max(start, ei->addr);
499 final_end = min(start + size, ei->addr + ei->size);
500 if (final_start >= final_end)
501 continue;
502 real_removed_size += final_end - final_start;
504 ei->size -= final_end - final_start;
505 if (ei->addr < final_start)
506 continue;
507 ei->addr = final_end;
509 return real_removed_size;
512 void __init update_e820(void)
514 int nr_map;
516 nr_map = e820.nr_map;
517 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
518 return;
519 e820.nr_map = nr_map;
520 printk(KERN_INFO "modified physical RAM map:\n");
521 e820_print_map("modified");
523 static void __init update_e820_saved(void)
525 int nr_map;
527 nr_map = e820_saved.nr_map;
528 if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
529 return;
530 e820_saved.nr_map = nr_map;
532 #define MAX_GAP_END 0x100000000ull
534 * Search for a gap in the e820 memory space from start_addr to end_addr.
536 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
537 unsigned long start_addr, unsigned long long end_addr)
539 unsigned long long last;
540 int i = e820.nr_map;
541 int found = 0;
543 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
545 while (--i >= 0) {
546 unsigned long long start = e820.map[i].addr;
547 unsigned long long end = start + e820.map[i].size;
549 if (end < start_addr)
550 continue;
553 * Since "last" is at most 4GB, we know we'll
554 * fit in 32 bits if this condition is true
556 if (last > end) {
557 unsigned long gap = last - end;
559 if (gap >= *gapsize) {
560 *gapsize = gap;
561 *gapstart = end;
562 found = 1;
565 if (start < last)
566 last = start;
568 return found;
572 * Search for the biggest gap in the low 32 bits of the e820
573 * memory space. We pass this space to PCI to assign MMIO resources
574 * for hotplug or unconfigured devices in.
575 * Hopefully the BIOS let enough space left.
577 __init void e820_setup_gap(void)
579 unsigned long gapstart, gapsize, round;
580 int found;
582 gapstart = 0x10000000;
583 gapsize = 0x400000;
584 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
586 #ifdef CONFIG_X86_64
587 if (!found) {
588 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
589 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
590 "address range\n"
591 KERN_ERR "PCI: Unassigned devices with 32bit resource "
592 "registers may break!\n");
594 #endif
597 * See how much we want to round up: start off with
598 * rounding to the next 1MB area.
600 round = 0x100000;
601 while ((gapsize >> 4) > round)
602 round += round;
603 /* Fun with two's complement */
604 pci_mem_start = (gapstart + round) & -round;
606 printk(KERN_INFO
607 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
608 pci_mem_start, gapstart, gapsize);
612 * Because of the size limitation of struct boot_params, only first
613 * 128 E820 memory entries are passed to kernel via
614 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
615 * linked list of struct setup_data, which is parsed here.
617 void __init parse_e820_ext(struct setup_data *sdata, unsigned long pa_data)
619 u32 map_len;
620 int entries;
621 struct e820entry *extmap;
623 entries = sdata->len / sizeof(struct e820entry);
624 map_len = sdata->len + sizeof(struct setup_data);
625 if (map_len > PAGE_SIZE)
626 sdata = early_ioremap(pa_data, map_len);
627 extmap = (struct e820entry *)(sdata->data);
628 __append_e820_map(extmap, entries);
629 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
630 if (map_len > PAGE_SIZE)
631 early_iounmap(sdata, map_len);
632 printk(KERN_INFO "extended physical RAM map:\n");
633 e820_print_map("extended");
636 #if defined(CONFIG_X86_64) || \
637 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
639 * Find the ranges of physical addresses that do not correspond to
640 * e820 RAM areas and mark the corresponding pages as nosave for
641 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
643 * This function requires the e820 map to be sorted and without any
644 * overlapping entries and assumes the first e820 area to be RAM.
646 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
648 int i;
649 unsigned long pfn;
651 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
652 for (i = 1; i < e820.nr_map; i++) {
653 struct e820entry *ei = &e820.map[i];
655 if (pfn < PFN_UP(ei->addr))
656 register_nosave_region(pfn, PFN_UP(ei->addr));
658 pfn = PFN_DOWN(ei->addr + ei->size);
659 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
660 register_nosave_region(PFN_UP(ei->addr), pfn);
662 if (pfn >= limit_pfn)
663 break;
666 #endif
669 * Early reserved memory areas.
671 #define MAX_EARLY_RES 20
673 struct early_res {
674 u64 start, end;
675 char name[16];
676 char overlap_ok;
678 static struct early_res early_res[MAX_EARLY_RES] __initdata = {
679 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
683 static int __init find_overlapped_early(u64 start, u64 end)
685 int i;
686 struct early_res *r;
688 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
689 r = &early_res[i];
690 if (end > r->start && start < r->end)
691 break;
694 return i;
698 * Drop the i-th range from the early reservation map,
699 * by copying any higher ranges down one over it, and
700 * clearing what had been the last slot.
702 static void __init drop_range(int i)
704 int j;
706 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
709 memmove(&early_res[i], &early_res[i + 1],
710 (j - 1 - i) * sizeof(struct early_res));
712 early_res[j - 1].end = 0;
716 * Split any existing ranges that:
717 * 1) are marked 'overlap_ok', and
718 * 2) overlap with the stated range [start, end)
719 * into whatever portion (if any) of the existing range is entirely
720 * below or entirely above the stated range. Drop the portion
721 * of the existing range that overlaps with the stated range,
722 * which will allow the caller of this routine to then add that
723 * stated range without conflicting with any existing range.
725 static void __init drop_overlaps_that_are_ok(u64 start, u64 end)
727 int i;
728 struct early_res *r;
729 u64 lower_start, lower_end;
730 u64 upper_start, upper_end;
731 char name[16];
733 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
734 r = &early_res[i];
736 /* Continue past non-overlapping ranges */
737 if (end <= r->start || start >= r->end)
738 continue;
741 * Leave non-ok overlaps as is; let caller
742 * panic "Overlapping early reservations"
743 * when it hits this overlap.
745 if (!r->overlap_ok)
746 return;
749 * We have an ok overlap. We will drop it from the early
750 * reservation map, and add back in any non-overlapping
751 * portions (lower or upper) as separate, overlap_ok,
752 * non-overlapping ranges.
755 /* 1. Note any non-overlapping (lower or upper) ranges. */
756 strncpy(name, r->name, sizeof(name) - 1);
758 lower_start = lower_end = 0;
759 upper_start = upper_end = 0;
760 if (r->start < start) {
761 lower_start = r->start;
762 lower_end = start;
764 if (r->end > end) {
765 upper_start = end;
766 upper_end = r->end;
769 /* 2. Drop the original ok overlapping range */
770 drop_range(i);
772 i--; /* resume for-loop on copied down entry */
774 /* 3. Add back in any non-overlapping ranges. */
775 if (lower_end)
776 reserve_early_overlap_ok(lower_start, lower_end, name);
777 if (upper_end)
778 reserve_early_overlap_ok(upper_start, upper_end, name);
782 static void __init __reserve_early(u64 start, u64 end, char *name,
783 int overlap_ok)
785 int i;
786 struct early_res *r;
788 i = find_overlapped_early(start, end);
789 if (i >= MAX_EARLY_RES)
790 panic("Too many early reservations");
791 r = &early_res[i];
792 if (r->end)
793 panic("Overlapping early reservations "
794 "%llx-%llx %s to %llx-%llx %s\n",
795 start, end - 1, name?name:"", r->start,
796 r->end - 1, r->name);
797 r->start = start;
798 r->end = end;
799 r->overlap_ok = overlap_ok;
800 if (name)
801 strncpy(r->name, name, sizeof(r->name) - 1);
805 * A few early reservtations come here.
807 * The 'overlap_ok' in the name of this routine does -not- mean it
808 * is ok for these reservations to overlap an earlier reservation.
809 * Rather it means that it is ok for subsequent reservations to
810 * overlap this one.
812 * Use this entry point to reserve early ranges when you are doing
813 * so out of "Paranoia", reserving perhaps more memory than you need,
814 * just in case, and don't mind a subsequent overlapping reservation
815 * that is known to be needed.
817 * The drop_overlaps_that_are_ok() call here isn't really needed.
818 * It would be needed if we had two colliding 'overlap_ok'
819 * reservations, so that the second such would not panic on the
820 * overlap with the first. We don't have any such as of this
821 * writing, but might as well tolerate such if it happens in
822 * the future.
824 void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
826 drop_overlaps_that_are_ok(start, end);
827 __reserve_early(start, end, name, 1);
831 * Most early reservations come here.
833 * We first have drop_overlaps_that_are_ok() drop any pre-existing
834 * 'overlap_ok' ranges, so that we can then reserve this memory
835 * range without risk of panic'ing on an overlapping overlap_ok
836 * early reservation.
838 void __init reserve_early(u64 start, u64 end, char *name)
840 drop_overlaps_that_are_ok(start, end);
841 __reserve_early(start, end, name, 0);
844 void __init free_early(u64 start, u64 end)
846 struct early_res *r;
847 int i;
849 i = find_overlapped_early(start, end);
850 r = &early_res[i];
851 if (i >= MAX_EARLY_RES || r->end != end || r->start != start)
852 panic("free_early on not reserved area: %llx-%llx!",
853 start, end - 1);
855 drop_range(i);
858 void __init early_res_to_bootmem(u64 start, u64 end)
860 int i, count;
861 u64 final_start, final_end;
863 count = 0;
864 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++)
865 count++;
867 printk(KERN_INFO "(%d early reservations) ==> bootmem [%010llx - %010llx]\n",
868 count, start, end);
869 for (i = 0; i < count; i++) {
870 struct early_res *r = &early_res[i];
871 printk(KERN_INFO " #%d [%010llx - %010llx] %16s", i,
872 r->start, r->end, r->name);
873 final_start = max(start, r->start);
874 final_end = min(end, r->end);
875 if (final_start >= final_end) {
876 printk(KERN_CONT "\n");
877 continue;
879 printk(KERN_CONT " ==> [%010llx - %010llx]\n",
880 final_start, final_end);
881 reserve_bootmem_generic(final_start, final_end - final_start,
882 BOOTMEM_DEFAULT);
886 /* Check for already reserved areas */
887 static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
889 int i;
890 u64 addr = *addrp;
891 int changed = 0;
892 struct early_res *r;
893 again:
894 i = find_overlapped_early(addr, addr + size);
895 r = &early_res[i];
896 if (i < MAX_EARLY_RES && r->end) {
897 *addrp = addr = round_up(r->end, align);
898 changed = 1;
899 goto again;
901 return changed;
904 /* Check for already reserved areas */
905 static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
907 int i;
908 u64 addr = *addrp, last;
909 u64 size = *sizep;
910 int changed = 0;
911 again:
912 last = addr + size;
913 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
914 struct early_res *r = &early_res[i];
915 if (last > r->start && addr < r->start) {
916 size = r->start - addr;
917 changed = 1;
918 goto again;
920 if (last > r->end && addr < r->end) {
921 addr = round_up(r->end, align);
922 size = last - addr;
923 changed = 1;
924 goto again;
926 if (last <= r->end && addr >= r->start) {
927 (*sizep)++;
928 return 0;
931 if (changed) {
932 *addrp = addr;
933 *sizep = size;
935 return changed;
939 * Find a free area with specified alignment in a specific range.
941 u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
943 int i;
945 for (i = 0; i < e820.nr_map; i++) {
946 struct e820entry *ei = &e820.map[i];
947 u64 addr, last;
948 u64 ei_last;
950 if (ei->type != E820_RAM)
951 continue;
952 addr = round_up(ei->addr, align);
953 ei_last = ei->addr + ei->size;
954 if (addr < start)
955 addr = round_up(start, align);
956 if (addr >= ei_last)
957 continue;
958 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
960 last = addr + size;
961 if (last > ei_last)
962 continue;
963 if (last > end)
964 continue;
965 return addr;
967 return -1ULL;
971 * Find next free range after *start
973 u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
975 int i;
977 for (i = 0; i < e820.nr_map; i++) {
978 struct e820entry *ei = &e820.map[i];
979 u64 addr, last;
980 u64 ei_last;
982 if (ei->type != E820_RAM)
983 continue;
984 addr = round_up(ei->addr, align);
985 ei_last = ei->addr + ei->size;
986 if (addr < start)
987 addr = round_up(start, align);
988 if (addr >= ei_last)
989 continue;
990 *sizep = ei_last - addr;
991 while (bad_addr_size(&addr, sizep, align) &&
992 addr + *sizep <= ei_last)
994 last = addr + *sizep;
995 if (last > ei_last)
996 continue;
997 return addr;
999 return -1UL;
1004 * pre allocated 4k and reserved it in e820
1006 u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
1008 u64 size = 0;
1009 u64 addr;
1010 u64 start;
1012 start = startt;
1013 while (size < sizet)
1014 start = find_e820_area_size(start, &size, align);
1016 if (size < sizet)
1017 return 0;
1019 addr = round_down(start + size - sizet, align);
1020 e820_update_range(addr, sizet, E820_RAM, E820_RESERVED);
1021 e820_update_range_saved(addr, sizet, E820_RAM, E820_RESERVED);
1022 printk(KERN_INFO "update e820 for early_reserve_e820\n");
1023 update_e820();
1024 update_e820_saved();
1026 return addr;
1029 #ifdef CONFIG_X86_32
1030 # ifdef CONFIG_X86_PAE
1031 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
1032 # else
1033 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
1034 # endif
1035 #else /* CONFIG_X86_32 */
1036 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
1037 #endif
1040 * Find the highest page frame number we have available
1042 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
1044 int i;
1045 unsigned long last_pfn = 0;
1046 unsigned long max_arch_pfn = MAX_ARCH_PFN;
1048 for (i = 0; i < e820.nr_map; i++) {
1049 struct e820entry *ei = &e820.map[i];
1050 unsigned long start_pfn;
1051 unsigned long end_pfn;
1053 if (ei->type != type)
1054 continue;
1056 start_pfn = ei->addr >> PAGE_SHIFT;
1057 end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
1059 if (start_pfn >= limit_pfn)
1060 continue;
1061 if (end_pfn > limit_pfn) {
1062 last_pfn = limit_pfn;
1063 break;
1065 if (end_pfn > last_pfn)
1066 last_pfn = end_pfn;
1069 if (last_pfn > max_arch_pfn)
1070 last_pfn = max_arch_pfn;
1072 printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
1073 last_pfn, max_arch_pfn);
1074 return last_pfn;
1076 unsigned long __init e820_end_of_ram_pfn(void)
1078 return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
1081 unsigned long __init e820_end_of_low_ram_pfn(void)
1083 return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
1086 * Finds an active region in the address range from start_pfn to last_pfn and
1087 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
1089 int __init e820_find_active_region(const struct e820entry *ei,
1090 unsigned long start_pfn,
1091 unsigned long last_pfn,
1092 unsigned long *ei_startpfn,
1093 unsigned long *ei_endpfn)
1095 u64 align = PAGE_SIZE;
1097 *ei_startpfn = round_up(ei->addr, align) >> PAGE_SHIFT;
1098 *ei_endpfn = round_down(ei->addr + ei->size, align) >> PAGE_SHIFT;
1100 /* Skip map entries smaller than a page */
1101 if (*ei_startpfn >= *ei_endpfn)
1102 return 0;
1104 /* Skip if map is outside the node */
1105 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
1106 *ei_startpfn >= last_pfn)
1107 return 0;
1109 /* Check for overlaps */
1110 if (*ei_startpfn < start_pfn)
1111 *ei_startpfn = start_pfn;
1112 if (*ei_endpfn > last_pfn)
1113 *ei_endpfn = last_pfn;
1115 return 1;
1118 /* Walk the e820 map and register active regions within a node */
1119 void __init e820_register_active_regions(int nid, unsigned long start_pfn,
1120 unsigned long last_pfn)
1122 unsigned long ei_startpfn;
1123 unsigned long ei_endpfn;
1124 int i;
1126 for (i = 0; i < e820.nr_map; i++)
1127 if (e820_find_active_region(&e820.map[i],
1128 start_pfn, last_pfn,
1129 &ei_startpfn, &ei_endpfn))
1130 add_active_range(nid, ei_startpfn, ei_endpfn);
1134 * Find the hole size (in bytes) in the memory range.
1135 * @start: starting address of the memory range to scan
1136 * @end: ending address of the memory range to scan
1138 u64 __init e820_hole_size(u64 start, u64 end)
1140 unsigned long start_pfn = start >> PAGE_SHIFT;
1141 unsigned long last_pfn = end >> PAGE_SHIFT;
1142 unsigned long ei_startpfn, ei_endpfn, ram = 0;
1143 int i;
1145 for (i = 0; i < e820.nr_map; i++) {
1146 if (e820_find_active_region(&e820.map[i],
1147 start_pfn, last_pfn,
1148 &ei_startpfn, &ei_endpfn))
1149 ram += ei_endpfn - ei_startpfn;
1151 return end - start - ((u64)ram << PAGE_SHIFT);
1154 static void early_panic(char *msg)
1156 early_printk(msg);
1157 panic(msg);
1160 static int userdef __initdata;
1162 /* "mem=nopentium" disables the 4MB page tables. */
1163 static int __init parse_memopt(char *p)
1165 u64 mem_size;
1167 if (!p)
1168 return -EINVAL;
1170 #ifdef CONFIG_X86_32
1171 if (!strcmp(p, "nopentium")) {
1172 setup_clear_cpu_cap(X86_FEATURE_PSE);
1173 return 0;
1175 #endif
1177 userdef = 1;
1178 mem_size = memparse(p, &p);
1179 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1181 return 0;
1183 early_param("mem", parse_memopt);
1185 static int __init parse_memmap_opt(char *p)
1187 char *oldp;
1188 u64 start_at, mem_size;
1190 if (!p)
1191 return -EINVAL;
1193 if (!strncmp(p, "exactmap", 8)) {
1194 #ifdef CONFIG_CRASH_DUMP
1196 * If we are doing a crash dump, we still need to know
1197 * the real mem size before original memory map is
1198 * reset.
1200 saved_max_pfn = e820_end_of_ram_pfn();
1201 #endif
1202 e820.nr_map = 0;
1203 userdef = 1;
1204 return 0;
1207 oldp = p;
1208 mem_size = memparse(p, &p);
1209 if (p == oldp)
1210 return -EINVAL;
1212 userdef = 1;
1213 if (*p == '@') {
1214 start_at = memparse(p+1, &p);
1215 e820_add_region(start_at, mem_size, E820_RAM);
1216 } else if (*p == '#') {
1217 start_at = memparse(p+1, &p);
1218 e820_add_region(start_at, mem_size, E820_ACPI);
1219 } else if (*p == '$') {
1220 start_at = memparse(p+1, &p);
1221 e820_add_region(start_at, mem_size, E820_RESERVED);
1222 } else
1223 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1225 return *p == '\0' ? 0 : -EINVAL;
1227 early_param("memmap", parse_memmap_opt);
1229 void __init finish_e820_parsing(void)
1231 if (userdef) {
1232 int nr = e820.nr_map;
1234 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
1235 early_panic("Invalid user supplied memory map");
1236 e820.nr_map = nr;
1238 printk(KERN_INFO "user-defined physical RAM map:\n");
1239 e820_print_map("user");
1243 static inline const char *e820_type_to_string(int e820_type)
1245 switch (e820_type) {
1246 case E820_RESERVED_KERN:
1247 case E820_RAM: return "System RAM";
1248 case E820_ACPI: return "ACPI Tables";
1249 case E820_NVS: return "ACPI Non-volatile Storage";
1250 case E820_UNUSABLE: return "Unusable memory";
1251 default: return "reserved";
1256 * Mark e820 reserved areas as busy for the resource manager.
1258 static struct resource __initdata *e820_res;
1259 void __init e820_reserve_resources(void)
1261 int i;
1262 struct resource *res;
1263 u64 end;
1265 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
1266 e820_res = res;
1267 for (i = 0; i < e820.nr_map; i++) {
1268 end = e820.map[i].addr + e820.map[i].size - 1;
1269 if (end != (resource_size_t)end) {
1270 res++;
1271 continue;
1273 res->name = e820_type_to_string(e820.map[i].type);
1274 res->start = e820.map[i].addr;
1275 res->end = end;
1277 res->flags = IORESOURCE_MEM;
1280 * don't register the region that could be conflicted with
1281 * pci device BAR resource and insert them later in
1282 * pcibios_resource_survey()
1284 if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20)) {
1285 res->flags |= IORESOURCE_BUSY;
1286 insert_resource(&iomem_resource, res);
1288 res++;
1291 for (i = 0; i < e820_saved.nr_map; i++) {
1292 struct e820entry *entry = &e820_saved.map[i];
1293 firmware_map_add_early(entry->addr,
1294 entry->addr + entry->size - 1,
1295 e820_type_to_string(entry->type));
1299 void __init e820_reserve_resources_late(void)
1301 int i;
1302 struct resource *res;
1304 res = e820_res;
1305 for (i = 0; i < e820.nr_map; i++) {
1306 if (!res->parent && res->end)
1307 insert_resource_expand_to_fit(&iomem_resource, res);
1308 res++;
1312 char *__init default_machine_specific_memory_setup(void)
1314 char *who = "BIOS-e820";
1315 int new_nr;
1317 * Try to copy the BIOS-supplied E820-map.
1319 * Otherwise fake a memory map; one section from 0k->640k,
1320 * the next section from 1mb->appropriate_mem_k
1322 new_nr = boot_params.e820_entries;
1323 sanitize_e820_map(boot_params.e820_map,
1324 ARRAY_SIZE(boot_params.e820_map),
1325 &new_nr);
1326 boot_params.e820_entries = new_nr;
1327 if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1328 < 0) {
1329 u64 mem_size;
1331 /* compare results from other methods and take the greater */
1332 if (boot_params.alt_mem_k
1333 < boot_params.screen_info.ext_mem_k) {
1334 mem_size = boot_params.screen_info.ext_mem_k;
1335 who = "BIOS-88";
1336 } else {
1337 mem_size = boot_params.alt_mem_k;
1338 who = "BIOS-e801";
1341 e820.nr_map = 0;
1342 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1343 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1346 /* In case someone cares... */
1347 return who;
1350 char *__init __attribute__((weak)) machine_specific_memory_setup(void)
1352 if (x86_quirks->arch_memory_setup) {
1353 char *who = x86_quirks->arch_memory_setup();
1355 if (who)
1356 return who;
1358 return default_machine_specific_memory_setup();
1361 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
1362 char * __init __attribute__((weak)) memory_setup(void)
1364 return machine_specific_memory_setup();
1367 void __init setup_memory_map(void)
1369 char *who;
1371 who = memory_setup();
1372 memcpy(&e820_saved, &e820, sizeof(struct e820map));
1373 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1374 e820_print_map(who);