Loading...
1/*
2 * Handle the memory map.
3 * The functions here do the job until bootmem takes over.
4 *
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>
9 *
10 */
11#include <linux/kernel.h>
12#include <linux/types.h>
13#include <linux/init.h>
14#include <linux/crash_dump.h>
15#include <linux/bootmem.h>
16#include <linux/pfn.h>
17#include <linux/suspend.h>
18#include <linux/acpi.h>
19#include <linux/firmware-map.h>
20#include <linux/memblock.h>
21
22#include <asm/e820.h>
23#include <asm/proto.h>
24#include <asm/setup.h>
25
26/*
27 * The e820 map is the map that gets modified e.g. with command line parameters
28 * and that is also registered with modifications in the kernel resource tree
29 * with the iomem_resource as parent.
30 *
31 * The e820_saved is directly saved after the BIOS-provided memory map is
32 * copied. It doesn't get modified afterwards. It's registered for the
33 * /sys/firmware/memmap interface.
34 *
35 * That memory map is not modified and is used as base for kexec. The kexec'd
36 * kernel should get the same memory map as the firmware provides. Then the
37 * user can e.g. boot the original kernel with mem=1G while still booting the
38 * next kernel with full memory.
39 */
40struct e820map e820;
41struct e820map e820_saved;
42
43/* For PCI or other memory-mapped resources */
44unsigned long pci_mem_start = 0xaeedbabe;
45#ifdef CONFIG_PCI
46EXPORT_SYMBOL(pci_mem_start);
47#endif
48
49/*
50 * This function checks if any part of the range <start,end> is mapped
51 * with type.
52 */
53int
54e820_any_mapped(u64 start, u64 end, unsigned type)
55{
56 int i;
57
58 for (i = 0; i < e820.nr_map; i++) {
59 struct e820entry *ei = &e820.map[i];
60
61 if (type && ei->type != type)
62 continue;
63 if (ei->addr >= end || ei->addr + ei->size <= start)
64 continue;
65 return 1;
66 }
67 return 0;
68}
69EXPORT_SYMBOL_GPL(e820_any_mapped);
70
71/*
72 * This function checks if the entire range <start,end> is mapped with type.
73 *
74 * Note: this function only works correct if the e820 table is sorted and
75 * not-overlapping, which is the case
76 */
77int __init e820_all_mapped(u64 start, u64 end, unsigned type)
78{
79 int i;
80
81 for (i = 0; i < e820.nr_map; i++) {
82 struct e820entry *ei = &e820.map[i];
83
84 if (type && ei->type != type)
85 continue;
86 /* is the region (part) in overlap with the current region ?*/
87 if (ei->addr >= end || ei->addr + ei->size <= start)
88 continue;
89
90 /* if the region is at the beginning of <start,end> we move
91 * start to the end of the region since it's ok until there
92 */
93 if (ei->addr <= start)
94 start = ei->addr + ei->size;
95 /*
96 * if start is now at or beyond end, we're done, full
97 * coverage
98 */
99 if (start >= end)
100 return 1;
101 }
102 return 0;
103}
104
105/*
106 * Add a memory region to the kernel e820 map.
107 */
108static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
109 int type)
110{
111 int x = e820x->nr_map;
112
113 if (x >= ARRAY_SIZE(e820x->map)) {
114 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
115 return;
116 }
117
118 e820x->map[x].addr = start;
119 e820x->map[x].size = size;
120 e820x->map[x].type = type;
121 e820x->nr_map++;
122}
123
124void __init e820_add_region(u64 start, u64 size, int type)
125{
126 __e820_add_region(&e820, start, size, type);
127}
128
129static void __init e820_print_type(u32 type)
130{
131 switch (type) {
132 case E820_RAM:
133 case E820_RESERVED_KERN:
134 printk(KERN_CONT "(usable)");
135 break;
136 case E820_RESERVED:
137 printk(KERN_CONT "(reserved)");
138 break;
139 case E820_ACPI:
140 printk(KERN_CONT "(ACPI data)");
141 break;
142 case E820_NVS:
143 printk(KERN_CONT "(ACPI NVS)");
144 break;
145 case E820_UNUSABLE:
146 printk(KERN_CONT "(unusable)");
147 break;
148 default:
149 printk(KERN_CONT "type %u", type);
150 break;
151 }
152}
153
154void __init e820_print_map(char *who)
155{
156 int i;
157
158 for (i = 0; i < e820.nr_map; i++) {
159 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
160 (unsigned long long) e820.map[i].addr,
161 (unsigned long long)
162 (e820.map[i].addr + e820.map[i].size));
163 e820_print_type(e820.map[i].type);
164 printk(KERN_CONT "\n");
165 }
166}
167
168/*
169 * Sanitize the BIOS e820 map.
170 *
171 * Some e820 responses include overlapping entries. The following
172 * replaces the original e820 map with a new one, removing overlaps,
173 * and resolving conflicting memory types in favor of highest
174 * numbered type.
175 *
176 * The input parameter biosmap points to an array of 'struct
177 * e820entry' which on entry has elements in the range [0, *pnr_map)
178 * valid, and which has space for up to max_nr_map entries.
179 * On return, the resulting sanitized e820 map entries will be in
180 * overwritten in the same location, starting at biosmap.
181 *
182 * The integer pointed to by pnr_map must be valid on entry (the
183 * current number of valid entries located at biosmap) and will
184 * be updated on return, with the new number of valid entries
185 * (something no more than max_nr_map.)
186 *
187 * The return value from sanitize_e820_map() is zero if it
188 * successfully 'sanitized' the map entries passed in, and is -1
189 * if it did nothing, which can happen if either of (1) it was
190 * only passed one map entry, or (2) any of the input map entries
191 * were invalid (start + size < start, meaning that the size was
192 * so big the described memory range wrapped around through zero.)
193 *
194 * Visually we're performing the following
195 * (1,2,3,4 = memory types)...
196 *
197 * Sample memory map (w/overlaps):
198 * ____22__________________
199 * ______________________4_
200 * ____1111________________
201 * _44_____________________
202 * 11111111________________
203 * ____________________33__
204 * ___________44___________
205 * __________33333_________
206 * ______________22________
207 * ___________________2222_
208 * _________111111111______
209 * _____________________11_
210 * _________________4______
211 *
212 * Sanitized equivalent (no overlap):
213 * 1_______________________
214 * _44_____________________
215 * ___1____________________
216 * ____22__________________
217 * ______11________________
218 * _________1______________
219 * __________3_____________
220 * ___________44___________
221 * _____________33_________
222 * _______________2________
223 * ________________1_______
224 * _________________4______
225 * ___________________2____
226 * ____________________33__
227 * ______________________4_
228 */
229
230int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
231 u32 *pnr_map)
232{
233 struct change_member {
234 struct e820entry *pbios; /* pointer to original bios entry */
235 unsigned long long addr; /* address for this change point */
236 };
237 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
238 static struct change_member *change_point[2*E820_X_MAX] __initdata;
239 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
240 static struct e820entry new_bios[E820_X_MAX] __initdata;
241 struct change_member *change_tmp;
242 unsigned long current_type, last_type;
243 unsigned long long last_addr;
244 int chgidx, still_changing;
245 int overlap_entries;
246 int new_bios_entry;
247 int old_nr, new_nr, chg_nr;
248 int i;
249
250 /* if there's only one memory region, don't bother */
251 if (*pnr_map < 2)
252 return -1;
253
254 old_nr = *pnr_map;
255 BUG_ON(old_nr > max_nr_map);
256
257 /* bail out if we find any unreasonable addresses in bios map */
258 for (i = 0; i < old_nr; i++)
259 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
260 return -1;
261
262 /* create pointers for initial change-point information (for sorting) */
263 for (i = 0; i < 2 * old_nr; i++)
264 change_point[i] = &change_point_list[i];
265
266 /* record all known change-points (starting and ending addresses),
267 omitting those that are for empty memory regions */
268 chgidx = 0;
269 for (i = 0; i < old_nr; i++) {
270 if (biosmap[i].size != 0) {
271 change_point[chgidx]->addr = biosmap[i].addr;
272 change_point[chgidx++]->pbios = &biosmap[i];
273 change_point[chgidx]->addr = biosmap[i].addr +
274 biosmap[i].size;
275 change_point[chgidx++]->pbios = &biosmap[i];
276 }
277 }
278 chg_nr = chgidx;
279
280 /* sort change-point list by memory addresses (low -> high) */
281 still_changing = 1;
282 while (still_changing) {
283 still_changing = 0;
284 for (i = 1; i < chg_nr; i++) {
285 unsigned long long curaddr, lastaddr;
286 unsigned long long curpbaddr, lastpbaddr;
287
288 curaddr = change_point[i]->addr;
289 lastaddr = change_point[i - 1]->addr;
290 curpbaddr = change_point[i]->pbios->addr;
291 lastpbaddr = change_point[i - 1]->pbios->addr;
292
293 /*
294 * swap entries, when:
295 *
296 * curaddr > lastaddr or
297 * curaddr == lastaddr and curaddr == curpbaddr and
298 * lastaddr != lastpbaddr
299 */
300 if (curaddr < lastaddr ||
301 (curaddr == lastaddr && curaddr == curpbaddr &&
302 lastaddr != lastpbaddr)) {
303 change_tmp = change_point[i];
304 change_point[i] = change_point[i-1];
305 change_point[i-1] = change_tmp;
306 still_changing = 1;
307 }
308 }
309 }
310
311 /* create a new bios memory map, removing overlaps */
312 overlap_entries = 0; /* number of entries in the overlap table */
313 new_bios_entry = 0; /* index for creating new bios map entries */
314 last_type = 0; /* start with undefined memory type */
315 last_addr = 0; /* start with 0 as last starting address */
316
317 /* loop through change-points, determining affect on the new bios map */
318 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
319 /* keep track of all overlapping bios entries */
320 if (change_point[chgidx]->addr ==
321 change_point[chgidx]->pbios->addr) {
322 /*
323 * add map entry to overlap list (> 1 entry
324 * implies an overlap)
325 */
326 overlap_list[overlap_entries++] =
327 change_point[chgidx]->pbios;
328 } else {
329 /*
330 * remove entry from list (order independent,
331 * so swap with last)
332 */
333 for (i = 0; i < overlap_entries; i++) {
334 if (overlap_list[i] ==
335 change_point[chgidx]->pbios)
336 overlap_list[i] =
337 overlap_list[overlap_entries-1];
338 }
339 overlap_entries--;
340 }
341 /*
342 * if there are overlapping entries, decide which
343 * "type" to use (larger value takes precedence --
344 * 1=usable, 2,3,4,4+=unusable)
345 */
346 current_type = 0;
347 for (i = 0; i < overlap_entries; i++)
348 if (overlap_list[i]->type > current_type)
349 current_type = overlap_list[i]->type;
350 /*
351 * continue building up new bios map based on this
352 * information
353 */
354 if (current_type != last_type) {
355 if (last_type != 0) {
356 new_bios[new_bios_entry].size =
357 change_point[chgidx]->addr - last_addr;
358 /*
359 * move forward only if the new size
360 * was non-zero
361 */
362 if (new_bios[new_bios_entry].size != 0)
363 /*
364 * no more space left for new
365 * bios entries ?
366 */
367 if (++new_bios_entry >= max_nr_map)
368 break;
369 }
370 if (current_type != 0) {
371 new_bios[new_bios_entry].addr =
372 change_point[chgidx]->addr;
373 new_bios[new_bios_entry].type = current_type;
374 last_addr = change_point[chgidx]->addr;
375 }
376 last_type = current_type;
377 }
378 }
379 /* retain count for new bios entries */
380 new_nr = new_bios_entry;
381
382 /* copy new bios mapping into original location */
383 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
384 *pnr_map = new_nr;
385
386 return 0;
387}
388
389static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
390{
391 while (nr_map) {
392 u64 start = biosmap->addr;
393 u64 size = biosmap->size;
394 u64 end = start + size;
395 u32 type = biosmap->type;
396
397 /* Overflow in 64 bits? Ignore the memory map. */
398 if (start > end)
399 return -1;
400
401 e820_add_region(start, size, type);
402
403 biosmap++;
404 nr_map--;
405 }
406 return 0;
407}
408
409/*
410 * Copy the BIOS e820 map into a safe place.
411 *
412 * Sanity-check it while we're at it..
413 *
414 * If we're lucky and live on a modern system, the setup code
415 * will have given us a memory map that we can use to properly
416 * set up memory. If we aren't, we'll fake a memory map.
417 */
418static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
419{
420 /* Only one memory region (or negative)? Ignore it */
421 if (nr_map < 2)
422 return -1;
423
424 return __append_e820_map(biosmap, nr_map);
425}
426
427static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
428 u64 size, unsigned old_type,
429 unsigned new_type)
430{
431 u64 end;
432 unsigned int i;
433 u64 real_updated_size = 0;
434
435 BUG_ON(old_type == new_type);
436
437 if (size > (ULLONG_MAX - start))
438 size = ULLONG_MAX - start;
439
440 end = start + size;
441 printk(KERN_DEBUG "e820 update range: %016Lx - %016Lx ",
442 (unsigned long long) start,
443 (unsigned long long) end);
444 e820_print_type(old_type);
445 printk(KERN_CONT " ==> ");
446 e820_print_type(new_type);
447 printk(KERN_CONT "\n");
448
449 for (i = 0; i < e820x->nr_map; i++) {
450 struct e820entry *ei = &e820x->map[i];
451 u64 final_start, final_end;
452 u64 ei_end;
453
454 if (ei->type != old_type)
455 continue;
456
457 ei_end = ei->addr + ei->size;
458 /* totally covered by new range? */
459 if (ei->addr >= start && ei_end <= end) {
460 ei->type = new_type;
461 real_updated_size += ei->size;
462 continue;
463 }
464
465 /* new range is totally covered? */
466 if (ei->addr < start && ei_end > end) {
467 __e820_add_region(e820x, start, size, new_type);
468 __e820_add_region(e820x, end, ei_end - end, ei->type);
469 ei->size = start - ei->addr;
470 real_updated_size += size;
471 continue;
472 }
473
474 /* partially covered */
475 final_start = max(start, ei->addr);
476 final_end = min(end, ei_end);
477 if (final_start >= final_end)
478 continue;
479
480 __e820_add_region(e820x, final_start, final_end - final_start,
481 new_type);
482
483 real_updated_size += final_end - final_start;
484
485 /*
486 * left range could be head or tail, so need to update
487 * size at first.
488 */
489 ei->size -= final_end - final_start;
490 if (ei->addr < final_start)
491 continue;
492 ei->addr = final_end;
493 }
494 return real_updated_size;
495}
496
497u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
498 unsigned new_type)
499{
500 return __e820_update_range(&e820, start, size, old_type, new_type);
501}
502
503static u64 __init e820_update_range_saved(u64 start, u64 size,
504 unsigned old_type, unsigned new_type)
505{
506 return __e820_update_range(&e820_saved, start, size, old_type,
507 new_type);
508}
509
510/* make e820 not cover the range */
511u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
512 int checktype)
513{
514 int i;
515 u64 end;
516 u64 real_removed_size = 0;
517
518 if (size > (ULLONG_MAX - start))
519 size = ULLONG_MAX - start;
520
521 end = start + size;
522 printk(KERN_DEBUG "e820 remove range: %016Lx - %016Lx ",
523 (unsigned long long) start,
524 (unsigned long long) end);
525 if (checktype)
526 e820_print_type(old_type);
527 printk(KERN_CONT "\n");
528
529 for (i = 0; i < e820.nr_map; i++) {
530 struct e820entry *ei = &e820.map[i];
531 u64 final_start, final_end;
532 u64 ei_end;
533
534 if (checktype && ei->type != old_type)
535 continue;
536
537 ei_end = ei->addr + ei->size;
538 /* totally covered? */
539 if (ei->addr >= start && ei_end <= end) {
540 real_removed_size += ei->size;
541 memset(ei, 0, sizeof(struct e820entry));
542 continue;
543 }
544
545 /* new range is totally covered? */
546 if (ei->addr < start && ei_end > end) {
547 e820_add_region(end, ei_end - end, ei->type);
548 ei->size = start - ei->addr;
549 real_removed_size += size;
550 continue;
551 }
552
553 /* partially covered */
554 final_start = max(start, ei->addr);
555 final_end = min(end, ei_end);
556 if (final_start >= final_end)
557 continue;
558 real_removed_size += final_end - final_start;
559
560 /*
561 * left range could be head or tail, so need to update
562 * size at first.
563 */
564 ei->size -= final_end - final_start;
565 if (ei->addr < final_start)
566 continue;
567 ei->addr = final_end;
568 }
569 return real_removed_size;
570}
571
572void __init update_e820(void)
573{
574 u32 nr_map;
575
576 nr_map = e820.nr_map;
577 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
578 return;
579 e820.nr_map = nr_map;
580 printk(KERN_INFO "modified physical RAM map:\n");
581 e820_print_map("modified");
582}
583static void __init update_e820_saved(void)
584{
585 u32 nr_map;
586
587 nr_map = e820_saved.nr_map;
588 if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
589 return;
590 e820_saved.nr_map = nr_map;
591}
592#define MAX_GAP_END 0x100000000ull
593/*
594 * Search for a gap in the e820 memory space from start_addr to end_addr.
595 */
596__init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
597 unsigned long start_addr, unsigned long long end_addr)
598{
599 unsigned long long last;
600 int i = e820.nr_map;
601 int found = 0;
602
603 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
604
605 while (--i >= 0) {
606 unsigned long long start = e820.map[i].addr;
607 unsigned long long end = start + e820.map[i].size;
608
609 if (end < start_addr)
610 continue;
611
612 /*
613 * Since "last" is at most 4GB, we know we'll
614 * fit in 32 bits if this condition is true
615 */
616 if (last > end) {
617 unsigned long gap = last - end;
618
619 if (gap >= *gapsize) {
620 *gapsize = gap;
621 *gapstart = end;
622 found = 1;
623 }
624 }
625 if (start < last)
626 last = start;
627 }
628 return found;
629}
630
631/*
632 * Search for the biggest gap in the low 32 bits of the e820
633 * memory space. We pass this space to PCI to assign MMIO resources
634 * for hotplug or unconfigured devices in.
635 * Hopefully the BIOS let enough space left.
636 */
637__init void e820_setup_gap(void)
638{
639 unsigned long gapstart, gapsize;
640 int found;
641
642 gapstart = 0x10000000;
643 gapsize = 0x400000;
644 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
645
646#ifdef CONFIG_X86_64
647 if (!found) {
648 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
649 printk(KERN_ERR
650 "PCI: Warning: Cannot find a gap in the 32bit address range\n"
651 "PCI: Unassigned devices with 32bit resource registers may break!\n");
652 }
653#endif
654
655 /*
656 * e820_reserve_resources_late protect stolen RAM already
657 */
658 pci_mem_start = gapstart;
659
660 printk(KERN_INFO
661 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
662 pci_mem_start, gapstart, gapsize);
663}
664
665/**
666 * Because of the size limitation of struct boot_params, only first
667 * 128 E820 memory entries are passed to kernel via
668 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
669 * linked list of struct setup_data, which is parsed here.
670 */
671void __init parse_e820_ext(struct setup_data *sdata)
672{
673 int entries;
674 struct e820entry *extmap;
675
676 entries = sdata->len / sizeof(struct e820entry);
677 extmap = (struct e820entry *)(sdata->data);
678 __append_e820_map(extmap, entries);
679 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
680 printk(KERN_INFO "extended physical RAM map:\n");
681 e820_print_map("extended");
682}
683
684#if defined(CONFIG_X86_64) || \
685 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
686/**
687 * Find the ranges of physical addresses that do not correspond to
688 * e820 RAM areas and mark the corresponding pages as nosave for
689 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
690 *
691 * This function requires the e820 map to be sorted and without any
692 * overlapping entries and assumes the first e820 area to be RAM.
693 */
694void __init e820_mark_nosave_regions(unsigned long limit_pfn)
695{
696 int i;
697 unsigned long pfn;
698
699 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
700 for (i = 1; i < e820.nr_map; i++) {
701 struct e820entry *ei = &e820.map[i];
702
703 if (pfn < PFN_UP(ei->addr))
704 register_nosave_region(pfn, PFN_UP(ei->addr));
705
706 pfn = PFN_DOWN(ei->addr + ei->size);
707 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
708 register_nosave_region(PFN_UP(ei->addr), pfn);
709
710 if (pfn >= limit_pfn)
711 break;
712 }
713}
714#endif
715
716#ifdef CONFIG_HIBERNATION
717/**
718 * Mark ACPI NVS memory region, so that we can save/restore it during
719 * hibernation and the subsequent resume.
720 */
721static int __init e820_mark_nvs_memory(void)
722{
723 int i;
724
725 for (i = 0; i < e820.nr_map; i++) {
726 struct e820entry *ei = &e820.map[i];
727
728 if (ei->type == E820_NVS)
729 suspend_nvs_register(ei->addr, ei->size);
730 }
731
732 return 0;
733}
734core_initcall(e820_mark_nvs_memory);
735#endif
736
737/*
738 * pre allocated 4k and reserved it in memblock and e820_saved
739 */
740u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
741{
742 u64 size = 0;
743 u64 addr;
744 u64 start;
745
746 for (start = startt; ; start += size) {
747 start = memblock_x86_find_in_range_size(start, &size, align);
748 if (start == MEMBLOCK_ERROR)
749 return 0;
750 if (size >= sizet)
751 break;
752 }
753
754#ifdef CONFIG_X86_32
755 if (start >= MAXMEM)
756 return 0;
757 if (start + size > MAXMEM)
758 size = MAXMEM - start;
759#endif
760
761 addr = round_down(start + size - sizet, align);
762 if (addr < start)
763 return 0;
764 memblock_x86_reserve_range(addr, addr + sizet, "new next");
765 e820_update_range_saved(addr, sizet, E820_RAM, E820_RESERVED);
766 printk(KERN_INFO "update e820_saved for early_reserve_e820\n");
767 update_e820_saved();
768
769 return addr;
770}
771
772#ifdef CONFIG_X86_32
773# ifdef CONFIG_X86_PAE
774# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
775# else
776# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
777# endif
778#else /* CONFIG_X86_32 */
779# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
780#endif
781
782/*
783 * Find the highest page frame number we have available
784 */
785static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
786{
787 int i;
788 unsigned long last_pfn = 0;
789 unsigned long max_arch_pfn = MAX_ARCH_PFN;
790
791 for (i = 0; i < e820.nr_map; i++) {
792 struct e820entry *ei = &e820.map[i];
793 unsigned long start_pfn;
794 unsigned long end_pfn;
795
796 if (ei->type != type)
797 continue;
798
799 start_pfn = ei->addr >> PAGE_SHIFT;
800 end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
801
802 if (start_pfn >= limit_pfn)
803 continue;
804 if (end_pfn > limit_pfn) {
805 last_pfn = limit_pfn;
806 break;
807 }
808 if (end_pfn > last_pfn)
809 last_pfn = end_pfn;
810 }
811
812 if (last_pfn > max_arch_pfn)
813 last_pfn = max_arch_pfn;
814
815 printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
816 last_pfn, max_arch_pfn);
817 return last_pfn;
818}
819unsigned long __init e820_end_of_ram_pfn(void)
820{
821 return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
822}
823
824unsigned long __init e820_end_of_low_ram_pfn(void)
825{
826 return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
827}
828
829static void early_panic(char *msg)
830{
831 early_printk(msg);
832 panic(msg);
833}
834
835static int userdef __initdata;
836
837/* "mem=nopentium" disables the 4MB page tables. */
838static int __init parse_memopt(char *p)
839{
840 u64 mem_size;
841
842 if (!p)
843 return -EINVAL;
844
845 if (!strcmp(p, "nopentium")) {
846#ifdef CONFIG_X86_32
847 setup_clear_cpu_cap(X86_FEATURE_PSE);
848 return 0;
849#else
850 printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
851 return -EINVAL;
852#endif
853 }
854
855 userdef = 1;
856 mem_size = memparse(p, &p);
857 /* don't remove all of memory when handling "mem={invalid}" param */
858 if (mem_size == 0)
859 return -EINVAL;
860 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
861
862 return 0;
863}
864early_param("mem", parse_memopt);
865
866static int __init parse_memmap_opt(char *p)
867{
868 char *oldp;
869 u64 start_at, mem_size;
870
871 if (!p)
872 return -EINVAL;
873
874 if (!strncmp(p, "exactmap", 8)) {
875#ifdef CONFIG_CRASH_DUMP
876 /*
877 * If we are doing a crash dump, we still need to know
878 * the real mem size before original memory map is
879 * reset.
880 */
881 saved_max_pfn = e820_end_of_ram_pfn();
882#endif
883 e820.nr_map = 0;
884 userdef = 1;
885 return 0;
886 }
887
888 oldp = p;
889 mem_size = memparse(p, &p);
890 if (p == oldp)
891 return -EINVAL;
892
893 userdef = 1;
894 if (*p == '@') {
895 start_at = memparse(p+1, &p);
896 e820_add_region(start_at, mem_size, E820_RAM);
897 } else if (*p == '#') {
898 start_at = memparse(p+1, &p);
899 e820_add_region(start_at, mem_size, E820_ACPI);
900 } else if (*p == '$') {
901 start_at = memparse(p+1, &p);
902 e820_add_region(start_at, mem_size, E820_RESERVED);
903 } else
904 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
905
906 return *p == '\0' ? 0 : -EINVAL;
907}
908early_param("memmap", parse_memmap_opt);
909
910void __init finish_e820_parsing(void)
911{
912 if (userdef) {
913 u32 nr = e820.nr_map;
914
915 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
916 early_panic("Invalid user supplied memory map");
917 e820.nr_map = nr;
918
919 printk(KERN_INFO "user-defined physical RAM map:\n");
920 e820_print_map("user");
921 }
922}
923
924static inline const char *e820_type_to_string(int e820_type)
925{
926 switch (e820_type) {
927 case E820_RESERVED_KERN:
928 case E820_RAM: return "System RAM";
929 case E820_ACPI: return "ACPI Tables";
930 case E820_NVS: return "ACPI Non-volatile Storage";
931 case E820_UNUSABLE: return "Unusable memory";
932 default: return "reserved";
933 }
934}
935
936/*
937 * Mark e820 reserved areas as busy for the resource manager.
938 */
939static struct resource __initdata *e820_res;
940void __init e820_reserve_resources(void)
941{
942 int i;
943 struct resource *res;
944 u64 end;
945
946 res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
947 e820_res = res;
948 for (i = 0; i < e820.nr_map; i++) {
949 end = e820.map[i].addr + e820.map[i].size - 1;
950 if (end != (resource_size_t)end) {
951 res++;
952 continue;
953 }
954 res->name = e820_type_to_string(e820.map[i].type);
955 res->start = e820.map[i].addr;
956 res->end = end;
957
958 res->flags = IORESOURCE_MEM;
959
960 /*
961 * don't register the region that could be conflicted with
962 * pci device BAR resource and insert them later in
963 * pcibios_resource_survey()
964 */
965 if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20)) {
966 res->flags |= IORESOURCE_BUSY;
967 insert_resource(&iomem_resource, res);
968 }
969 res++;
970 }
971
972 for (i = 0; i < e820_saved.nr_map; i++) {
973 struct e820entry *entry = &e820_saved.map[i];
974 firmware_map_add_early(entry->addr,
975 entry->addr + entry->size - 1,
976 e820_type_to_string(entry->type));
977 }
978}
979
980/* How much should we pad RAM ending depending on where it is? */
981static unsigned long ram_alignment(resource_size_t pos)
982{
983 unsigned long mb = pos >> 20;
984
985 /* To 64kB in the first megabyte */
986 if (!mb)
987 return 64*1024;
988
989 /* To 1MB in the first 16MB */
990 if (mb < 16)
991 return 1024*1024;
992
993 /* To 64MB for anything above that */
994 return 64*1024*1024;
995}
996
997#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
998
999void __init e820_reserve_resources_late(void)
1000{
1001 int i;
1002 struct resource *res;
1003
1004 res = e820_res;
1005 for (i = 0; i < e820.nr_map; i++) {
1006 if (!res->parent && res->end)
1007 insert_resource_expand_to_fit(&iomem_resource, res);
1008 res++;
1009 }
1010
1011 /*
1012 * Try to bump up RAM regions to reasonable boundaries to
1013 * avoid stolen RAM:
1014 */
1015 for (i = 0; i < e820.nr_map; i++) {
1016 struct e820entry *entry = &e820.map[i];
1017 u64 start, end;
1018
1019 if (entry->type != E820_RAM)
1020 continue;
1021 start = entry->addr + entry->size;
1022 end = round_up(start, ram_alignment(start)) - 1;
1023 if (end > MAX_RESOURCE_SIZE)
1024 end = MAX_RESOURCE_SIZE;
1025 if (start >= end)
1026 continue;
1027 printk(KERN_DEBUG "reserve RAM buffer: %016llx - %016llx ",
1028 start, end);
1029 reserve_region_with_split(&iomem_resource, start, end,
1030 "RAM buffer");
1031 }
1032}
1033
1034char *__init default_machine_specific_memory_setup(void)
1035{
1036 char *who = "BIOS-e820";
1037 u32 new_nr;
1038 /*
1039 * Try to copy the BIOS-supplied E820-map.
1040 *
1041 * Otherwise fake a memory map; one section from 0k->640k,
1042 * the next section from 1mb->appropriate_mem_k
1043 */
1044 new_nr = boot_params.e820_entries;
1045 sanitize_e820_map(boot_params.e820_map,
1046 ARRAY_SIZE(boot_params.e820_map),
1047 &new_nr);
1048 boot_params.e820_entries = new_nr;
1049 if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1050 < 0) {
1051 u64 mem_size;
1052
1053 /* compare results from other methods and take the greater */
1054 if (boot_params.alt_mem_k
1055 < boot_params.screen_info.ext_mem_k) {
1056 mem_size = boot_params.screen_info.ext_mem_k;
1057 who = "BIOS-88";
1058 } else {
1059 mem_size = boot_params.alt_mem_k;
1060 who = "BIOS-e801";
1061 }
1062
1063 e820.nr_map = 0;
1064 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1065 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1066 }
1067
1068 /* In case someone cares... */
1069 return who;
1070}
1071
1072void __init setup_memory_map(void)
1073{
1074 char *who;
1075
1076 who = x86_init.resources.memory_setup();
1077 memcpy(&e820_saved, &e820, sizeof(struct e820map));
1078 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1079 e820_print_map(who);
1080}
1081
1082void __init memblock_x86_fill(void)
1083{
1084 int i;
1085 u64 end;
1086
1087 /*
1088 * EFI may have more than 128 entries
1089 * We are safe to enable resizing, beause memblock_x86_fill()
1090 * is rather later for x86
1091 */
1092 memblock_can_resize = 1;
1093
1094 for (i = 0; i < e820.nr_map; i++) {
1095 struct e820entry *ei = &e820.map[i];
1096
1097 end = ei->addr + ei->size;
1098 if (end != (resource_size_t)end)
1099 continue;
1100
1101 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1102 continue;
1103
1104 memblock_add(ei->addr, ei->size);
1105 }
1106
1107 memblock_analyze();
1108 memblock_dump_all();
1109}
1110
1111void __init memblock_find_dma_reserve(void)
1112{
1113#ifdef CONFIG_X86_64
1114 u64 free_size_pfn;
1115 u64 mem_size_pfn;
1116 /*
1117 * need to find out used area below MAX_DMA_PFN
1118 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1119 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1120 */
1121 mem_size_pfn = memblock_x86_memory_in_range(0, MAX_DMA_PFN << PAGE_SHIFT) >> PAGE_SHIFT;
1122 free_size_pfn = memblock_x86_free_memory_in_range(0, MAX_DMA_PFN << PAGE_SHIFT) >> PAGE_SHIFT;
1123 set_dma_reserve(mem_size_pfn - free_size_pfn);
1124#endif
1125}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Low level x86 E820 memory map handling functions.
4 *
5 * The firmware and bootloader passes us the "E820 table", which is the primary
6 * physical memory layout description available about x86 systems.
7 *
8 * The kernel takes the E820 memory layout and optionally modifies it with
9 * quirks and other tweaks, and feeds that into the generic Linux memory
10 * allocation code routines via a platform independent interface (memblock, etc.).
11 */
12#include <linux/crash_dump.h>
13#include <linux/memblock.h>
14#include <linux/suspend.h>
15#include <linux/acpi.h>
16#include <linux/firmware-map.h>
17#include <linux/sort.h>
18#include <linux/memory_hotplug.h>
19
20#include <asm/e820/api.h>
21#include <asm/setup.h>
22
23/*
24 * We organize the E820 table into three main data structures:
25 *
26 * - 'e820_table_firmware': the original firmware version passed to us by the
27 * bootloader - not modified by the kernel. It is composed of two parts:
28 * the first 128 E820 memory entries in boot_params.e820_table and the remaining
29 * (if any) entries of the SETUP_E820_EXT nodes. We use this to:
30 *
31 * - inform the user about the firmware's notion of memory layout
32 * via /sys/firmware/memmap
33 *
34 * - the hibernation code uses it to generate a kernel-independent MD5
35 * fingerprint of the physical memory layout of a system.
36 *
37 * - 'e820_table_kexec': a slightly modified (by the kernel) firmware version
38 * passed to us by the bootloader - the major difference between
39 * e820_table_firmware[] and this one is that, the latter marks the setup_data
40 * list created by the EFI boot stub as reserved, so that kexec can reuse the
41 * setup_data information in the second kernel. Besides, e820_table_kexec[]
42 * might also be modified by the kexec itself to fake a mptable.
43 * We use this to:
44 *
45 * - kexec, which is a bootloader in disguise, uses the original E820
46 * layout to pass to the kexec-ed kernel. This way the original kernel
47 * can have a restricted E820 map while the kexec()-ed kexec-kernel
48 * can have access to full memory - etc.
49 *
50 * - 'e820_table': this is the main E820 table that is massaged by the
51 * low level x86 platform code, or modified by boot parameters, before
52 * passed on to higher level MM layers.
53 *
54 * Once the E820 map has been converted to the standard Linux memory layout
55 * information its role stops - modifying it has no effect and does not get
56 * re-propagated. So itsmain role is a temporary bootstrap storage of firmware
57 * specific memory layout data during early bootup.
58 */
59static struct e820_table e820_table_init __initdata;
60static struct e820_table e820_table_kexec_init __initdata;
61static struct e820_table e820_table_firmware_init __initdata;
62
63struct e820_table *e820_table __refdata = &e820_table_init;
64struct e820_table *e820_table_kexec __refdata = &e820_table_kexec_init;
65struct e820_table *e820_table_firmware __refdata = &e820_table_firmware_init;
66
67/* For PCI or other memory-mapped resources */
68unsigned long pci_mem_start = 0xaeedbabe;
69#ifdef CONFIG_PCI
70EXPORT_SYMBOL(pci_mem_start);
71#endif
72
73/*
74 * This function checks if any part of the range <start,end> is mapped
75 * with type.
76 */
77static bool _e820__mapped_any(struct e820_table *table,
78 u64 start, u64 end, enum e820_type type)
79{
80 int i;
81
82 for (i = 0; i < table->nr_entries; i++) {
83 struct e820_entry *entry = &table->entries[i];
84
85 if (type && entry->type != type)
86 continue;
87 if (entry->addr >= end || entry->addr + entry->size <= start)
88 continue;
89 return true;
90 }
91 return false;
92}
93
94bool e820__mapped_raw_any(u64 start, u64 end, enum e820_type type)
95{
96 return _e820__mapped_any(e820_table_firmware, start, end, type);
97}
98EXPORT_SYMBOL_GPL(e820__mapped_raw_any);
99
100bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
101{
102 return _e820__mapped_any(e820_table, start, end, type);
103}
104EXPORT_SYMBOL_GPL(e820__mapped_any);
105
106/*
107 * This function checks if the entire <start,end> range is mapped with 'type'.
108 *
109 * Note: this function only works correctly once the E820 table is sorted and
110 * not-overlapping (at least for the range specified), which is the case normally.
111 */
112static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
113 enum e820_type type)
114{
115 int i;
116
117 for (i = 0; i < e820_table->nr_entries; i++) {
118 struct e820_entry *entry = &e820_table->entries[i];
119
120 if (type && entry->type != type)
121 continue;
122
123 /* Is the region (part) in overlap with the current region? */
124 if (entry->addr >= end || entry->addr + entry->size <= start)
125 continue;
126
127 /*
128 * If the region is at the beginning of <start,end> we move
129 * 'start' to the end of the region since it's ok until there
130 */
131 if (entry->addr <= start)
132 start = entry->addr + entry->size;
133
134 /*
135 * If 'start' is now at or beyond 'end', we're done, full
136 * coverage of the desired range exists:
137 */
138 if (start >= end)
139 return entry;
140 }
141
142 return NULL;
143}
144
145/*
146 * This function checks if the entire range <start,end> is mapped with type.
147 */
148bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
149{
150 return __e820__mapped_all(start, end, type);
151}
152
153/*
154 * This function returns the type associated with the range <start,end>.
155 */
156int e820__get_entry_type(u64 start, u64 end)
157{
158 struct e820_entry *entry = __e820__mapped_all(start, end, 0);
159
160 return entry ? entry->type : -EINVAL;
161}
162
163/*
164 * Add a memory region to the kernel E820 map.
165 */
166static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
167{
168 int x = table->nr_entries;
169
170 if (x >= ARRAY_SIZE(table->entries)) {
171 pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
172 start, start + size - 1);
173 return;
174 }
175
176 table->entries[x].addr = start;
177 table->entries[x].size = size;
178 table->entries[x].type = type;
179 table->nr_entries++;
180}
181
182void __init e820__range_add(u64 start, u64 size, enum e820_type type)
183{
184 __e820__range_add(e820_table, start, size, type);
185}
186
187static void __init e820_print_type(enum e820_type type)
188{
189 switch (type) {
190 case E820_TYPE_RAM: /* Fall through: */
191 case E820_TYPE_RESERVED_KERN: pr_cont("usable"); break;
192 case E820_TYPE_RESERVED: pr_cont("reserved"); break;
193 case E820_TYPE_SOFT_RESERVED: pr_cont("soft reserved"); break;
194 case E820_TYPE_ACPI: pr_cont("ACPI data"); break;
195 case E820_TYPE_NVS: pr_cont("ACPI NVS"); break;
196 case E820_TYPE_UNUSABLE: pr_cont("unusable"); break;
197 case E820_TYPE_PMEM: /* Fall through: */
198 case E820_TYPE_PRAM: pr_cont("persistent (type %u)", type); break;
199 default: pr_cont("type %u", type); break;
200 }
201}
202
203void __init e820__print_table(char *who)
204{
205 int i;
206
207 for (i = 0; i < e820_table->nr_entries; i++) {
208 pr_info("%s: [mem %#018Lx-%#018Lx] ",
209 who,
210 e820_table->entries[i].addr,
211 e820_table->entries[i].addr + e820_table->entries[i].size - 1);
212
213 e820_print_type(e820_table->entries[i].type);
214 pr_cont("\n");
215 }
216}
217
218/*
219 * Sanitize an E820 map.
220 *
221 * Some E820 layouts include overlapping entries. The following
222 * replaces the original E820 map with a new one, removing overlaps,
223 * and resolving conflicting memory types in favor of highest
224 * numbered type.
225 *
226 * The input parameter 'entries' points to an array of 'struct
227 * e820_entry' which on entry has elements in the range [0, *nr_entries)
228 * valid, and which has space for up to max_nr_entries entries.
229 * On return, the resulting sanitized E820 map entries will be in
230 * overwritten in the same location, starting at 'entries'.
231 *
232 * The integer pointed to by nr_entries must be valid on entry (the
233 * current number of valid entries located at 'entries'). If the
234 * sanitizing succeeds the *nr_entries will be updated with the new
235 * number of valid entries (something no more than max_nr_entries).
236 *
237 * The return value from e820__update_table() is zero if it
238 * successfully 'sanitized' the map entries passed in, and is -1
239 * if it did nothing, which can happen if either of (1) it was
240 * only passed one map entry, or (2) any of the input map entries
241 * were invalid (start + size < start, meaning that the size was
242 * so big the described memory range wrapped around through zero.)
243 *
244 * Visually we're performing the following
245 * (1,2,3,4 = memory types)...
246 *
247 * Sample memory map (w/overlaps):
248 * ____22__________________
249 * ______________________4_
250 * ____1111________________
251 * _44_____________________
252 * 11111111________________
253 * ____________________33__
254 * ___________44___________
255 * __________33333_________
256 * ______________22________
257 * ___________________2222_
258 * _________111111111______
259 * _____________________11_
260 * _________________4______
261 *
262 * Sanitized equivalent (no overlap):
263 * 1_______________________
264 * _44_____________________
265 * ___1____________________
266 * ____22__________________
267 * ______11________________
268 * _________1______________
269 * __________3_____________
270 * ___________44___________
271 * _____________33_________
272 * _______________2________
273 * ________________1_______
274 * _________________4______
275 * ___________________2____
276 * ____________________33__
277 * ______________________4_
278 */
279struct change_member {
280 /* Pointer to the original entry: */
281 struct e820_entry *entry;
282 /* Address for this change point: */
283 unsigned long long addr;
284};
285
286static struct change_member change_point_list[2*E820_MAX_ENTRIES] __initdata;
287static struct change_member *change_point[2*E820_MAX_ENTRIES] __initdata;
288static struct e820_entry *overlap_list[E820_MAX_ENTRIES] __initdata;
289static struct e820_entry new_entries[E820_MAX_ENTRIES] __initdata;
290
291static int __init cpcompare(const void *a, const void *b)
292{
293 struct change_member * const *app = a, * const *bpp = b;
294 const struct change_member *ap = *app, *bp = *bpp;
295
296 /*
297 * Inputs are pointers to two elements of change_point[]. If their
298 * addresses are not equal, their difference dominates. If the addresses
299 * are equal, then consider one that represents the end of its region
300 * to be greater than one that does not.
301 */
302 if (ap->addr != bp->addr)
303 return ap->addr > bp->addr ? 1 : -1;
304
305 return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
306}
307
308int __init e820__update_table(struct e820_table *table)
309{
310 struct e820_entry *entries = table->entries;
311 u32 max_nr_entries = ARRAY_SIZE(table->entries);
312 enum e820_type current_type, last_type;
313 unsigned long long last_addr;
314 u32 new_nr_entries, overlap_entries;
315 u32 i, chg_idx, chg_nr;
316
317 /* If there's only one memory region, don't bother: */
318 if (table->nr_entries < 2)
319 return -1;
320
321 BUG_ON(table->nr_entries > max_nr_entries);
322
323 /* Bail out if we find any unreasonable addresses in the map: */
324 for (i = 0; i < table->nr_entries; i++) {
325 if (entries[i].addr + entries[i].size < entries[i].addr)
326 return -1;
327 }
328
329 /* Create pointers for initial change-point information (for sorting): */
330 for (i = 0; i < 2 * table->nr_entries; i++)
331 change_point[i] = &change_point_list[i];
332
333 /*
334 * Record all known change-points (starting and ending addresses),
335 * omitting empty memory regions:
336 */
337 chg_idx = 0;
338 for (i = 0; i < table->nr_entries; i++) {
339 if (entries[i].size != 0) {
340 change_point[chg_idx]->addr = entries[i].addr;
341 change_point[chg_idx++]->entry = &entries[i];
342 change_point[chg_idx]->addr = entries[i].addr + entries[i].size;
343 change_point[chg_idx++]->entry = &entries[i];
344 }
345 }
346 chg_nr = chg_idx;
347
348 /* Sort change-point list by memory addresses (low -> high): */
349 sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
350
351 /* Create a new memory map, removing overlaps: */
352 overlap_entries = 0; /* Number of entries in the overlap table */
353 new_nr_entries = 0; /* Index for creating new map entries */
354 last_type = 0; /* Start with undefined memory type */
355 last_addr = 0; /* Start with 0 as last starting address */
356
357 /* Loop through change-points, determining effect on the new map: */
358 for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
359 /* Keep track of all overlapping entries */
360 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
361 /* Add map entry to overlap list (> 1 entry implies an overlap) */
362 overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
363 } else {
364 /* Remove entry from list (order independent, so swap with last): */
365 for (i = 0; i < overlap_entries; i++) {
366 if (overlap_list[i] == change_point[chg_idx]->entry)
367 overlap_list[i] = overlap_list[overlap_entries-1];
368 }
369 overlap_entries--;
370 }
371 /*
372 * If there are overlapping entries, decide which
373 * "type" to use (larger value takes precedence --
374 * 1=usable, 2,3,4,4+=unusable)
375 */
376 current_type = 0;
377 for (i = 0; i < overlap_entries; i++) {
378 if (overlap_list[i]->type > current_type)
379 current_type = overlap_list[i]->type;
380 }
381
382 /* Continue building up new map based on this information: */
383 if (current_type != last_type || current_type == E820_TYPE_PRAM) {
384 if (last_type != 0) {
385 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
386 /* Move forward only if the new size was non-zero: */
387 if (new_entries[new_nr_entries].size != 0)
388 /* No more space left for new entries? */
389 if (++new_nr_entries >= max_nr_entries)
390 break;
391 }
392 if (current_type != 0) {
393 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
394 new_entries[new_nr_entries].type = current_type;
395 last_addr = change_point[chg_idx]->addr;
396 }
397 last_type = current_type;
398 }
399 }
400
401 /* Copy the new entries into the original location: */
402 memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
403 table->nr_entries = new_nr_entries;
404
405 return 0;
406}
407
408static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
409{
410 struct boot_e820_entry *entry = entries;
411
412 while (nr_entries) {
413 u64 start = entry->addr;
414 u64 size = entry->size;
415 u64 end = start + size - 1;
416 u32 type = entry->type;
417
418 /* Ignore the entry on 64-bit overflow: */
419 if (start > end && likely(size))
420 return -1;
421
422 e820__range_add(start, size, type);
423
424 entry++;
425 nr_entries--;
426 }
427 return 0;
428}
429
430/*
431 * Copy the BIOS E820 map into a safe place.
432 *
433 * Sanity-check it while we're at it..
434 *
435 * If we're lucky and live on a modern system, the setup code
436 * will have given us a memory map that we can use to properly
437 * set up memory. If we aren't, we'll fake a memory map.
438 */
439static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
440{
441 /* Only one memory region (or negative)? Ignore it */
442 if (nr_entries < 2)
443 return -1;
444
445 return __append_e820_table(entries, nr_entries);
446}
447
448static u64 __init
449__e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
450{
451 u64 end;
452 unsigned int i;
453 u64 real_updated_size = 0;
454
455 BUG_ON(old_type == new_type);
456
457 if (size > (ULLONG_MAX - start))
458 size = ULLONG_MAX - start;
459
460 end = start + size;
461 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
462 e820_print_type(old_type);
463 pr_cont(" ==> ");
464 e820_print_type(new_type);
465 pr_cont("\n");
466
467 for (i = 0; i < table->nr_entries; i++) {
468 struct e820_entry *entry = &table->entries[i];
469 u64 final_start, final_end;
470 u64 entry_end;
471
472 if (entry->type != old_type)
473 continue;
474
475 entry_end = entry->addr + entry->size;
476
477 /* Completely covered by new range? */
478 if (entry->addr >= start && entry_end <= end) {
479 entry->type = new_type;
480 real_updated_size += entry->size;
481 continue;
482 }
483
484 /* New range is completely covered? */
485 if (entry->addr < start && entry_end > end) {
486 __e820__range_add(table, start, size, new_type);
487 __e820__range_add(table, end, entry_end - end, entry->type);
488 entry->size = start - entry->addr;
489 real_updated_size += size;
490 continue;
491 }
492
493 /* Partially covered: */
494 final_start = max(start, entry->addr);
495 final_end = min(end, entry_end);
496 if (final_start >= final_end)
497 continue;
498
499 __e820__range_add(table, final_start, final_end - final_start, new_type);
500
501 real_updated_size += final_end - final_start;
502
503 /*
504 * Left range could be head or tail, so need to update
505 * its size first:
506 */
507 entry->size -= final_end - final_start;
508 if (entry->addr < final_start)
509 continue;
510
511 entry->addr = final_end;
512 }
513 return real_updated_size;
514}
515
516u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
517{
518 return __e820__range_update(e820_table, start, size, old_type, new_type);
519}
520
521static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
522{
523 return __e820__range_update(e820_table_kexec, start, size, old_type, new_type);
524}
525
526/* Remove a range of memory from the E820 table: */
527u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
528{
529 int i;
530 u64 end;
531 u64 real_removed_size = 0;
532
533 if (size > (ULLONG_MAX - start))
534 size = ULLONG_MAX - start;
535
536 end = start + size;
537 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
538 if (check_type)
539 e820_print_type(old_type);
540 pr_cont("\n");
541
542 for (i = 0; i < e820_table->nr_entries; i++) {
543 struct e820_entry *entry = &e820_table->entries[i];
544 u64 final_start, final_end;
545 u64 entry_end;
546
547 if (check_type && entry->type != old_type)
548 continue;
549
550 entry_end = entry->addr + entry->size;
551
552 /* Completely covered? */
553 if (entry->addr >= start && entry_end <= end) {
554 real_removed_size += entry->size;
555 memset(entry, 0, sizeof(*entry));
556 continue;
557 }
558
559 /* Is the new range completely covered? */
560 if (entry->addr < start && entry_end > end) {
561 e820__range_add(end, entry_end - end, entry->type);
562 entry->size = start - entry->addr;
563 real_removed_size += size;
564 continue;
565 }
566
567 /* Partially covered: */
568 final_start = max(start, entry->addr);
569 final_end = min(end, entry_end);
570 if (final_start >= final_end)
571 continue;
572
573 real_removed_size += final_end - final_start;
574
575 /*
576 * Left range could be head or tail, so need to update
577 * the size first:
578 */
579 entry->size -= final_end - final_start;
580 if (entry->addr < final_start)
581 continue;
582
583 entry->addr = final_end;
584 }
585 return real_removed_size;
586}
587
588void __init e820__update_table_print(void)
589{
590 if (e820__update_table(e820_table))
591 return;
592
593 pr_info("modified physical RAM map:\n");
594 e820__print_table("modified");
595}
596
597static void __init e820__update_table_kexec(void)
598{
599 e820__update_table(e820_table_kexec);
600}
601
602#define MAX_GAP_END 0x100000000ull
603
604/*
605 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
606 */
607static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
608{
609 unsigned long long last = MAX_GAP_END;
610 int i = e820_table->nr_entries;
611 int found = 0;
612
613 while (--i >= 0) {
614 unsigned long long start = e820_table->entries[i].addr;
615 unsigned long long end = start + e820_table->entries[i].size;
616
617 /*
618 * Since "last" is at most 4GB, we know we'll
619 * fit in 32 bits if this condition is true:
620 */
621 if (last > end) {
622 unsigned long gap = last - end;
623
624 if (gap >= *gapsize) {
625 *gapsize = gap;
626 *gapstart = end;
627 found = 1;
628 }
629 }
630 if (start < last)
631 last = start;
632 }
633 return found;
634}
635
636/*
637 * Search for the biggest gap in the low 32 bits of the E820
638 * memory space. We pass this space to the PCI subsystem, so
639 * that it can assign MMIO resources for hotplug or
640 * unconfigured devices in.
641 *
642 * Hopefully the BIOS let enough space left.
643 */
644__init void e820__setup_pci_gap(void)
645{
646 unsigned long gapstart, gapsize;
647 int found;
648
649 gapsize = 0x400000;
650 found = e820_search_gap(&gapstart, &gapsize);
651
652 if (!found) {
653#ifdef CONFIG_X86_64
654 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
655 pr_err("Cannot find an available gap in the 32-bit address range\n");
656 pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
657#else
658 gapstart = 0x10000000;
659#endif
660 }
661
662 /*
663 * e820__reserve_resources_late() protects stolen RAM already:
664 */
665 pci_mem_start = gapstart;
666
667 pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
668 gapstart, gapstart + gapsize - 1);
669}
670
671/*
672 * Called late during init, in free_initmem().
673 *
674 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
675 *
676 * Copy them to a (usually much smaller) dynamically allocated area that is
677 * sized precisely after the number of e820 entries.
678 *
679 * This is done after we've performed all the fixes and tweaks to the tables.
680 * All functions which modify them are __init functions, which won't exist
681 * after free_initmem().
682 */
683__init void e820__reallocate_tables(void)
684{
685 struct e820_table *n;
686 int size;
687
688 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
689 n = kmemdup(e820_table, size, GFP_KERNEL);
690 BUG_ON(!n);
691 e820_table = n;
692
693 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
694 n = kmemdup(e820_table_kexec, size, GFP_KERNEL);
695 BUG_ON(!n);
696 e820_table_kexec = n;
697
698 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
699 n = kmemdup(e820_table_firmware, size, GFP_KERNEL);
700 BUG_ON(!n);
701 e820_table_firmware = n;
702}
703
704/*
705 * Because of the small fixed size of struct boot_params, only the first
706 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
707 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
708 * struct setup_data, which is parsed here.
709 */
710void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
711{
712 int entries;
713 struct boot_e820_entry *extmap;
714 struct setup_data *sdata;
715
716 sdata = early_memremap(phys_addr, data_len);
717 entries = sdata->len / sizeof(*extmap);
718 extmap = (struct boot_e820_entry *)(sdata->data);
719
720 __append_e820_table(extmap, entries);
721 e820__update_table(e820_table);
722
723 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
724 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
725
726 early_memunmap(sdata, data_len);
727 pr_info("extended physical RAM map:\n");
728 e820__print_table("extended");
729}
730
731/*
732 * Find the ranges of physical addresses that do not correspond to
733 * E820 RAM areas and register the corresponding pages as 'nosave' for
734 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
735 *
736 * This function requires the E820 map to be sorted and without any
737 * overlapping entries.
738 */
739void __init e820__register_nosave_regions(unsigned long limit_pfn)
740{
741 int i;
742 unsigned long pfn = 0;
743
744 for (i = 0; i < e820_table->nr_entries; i++) {
745 struct e820_entry *entry = &e820_table->entries[i];
746
747 if (pfn < PFN_UP(entry->addr))
748 register_nosave_region(pfn, PFN_UP(entry->addr));
749
750 pfn = PFN_DOWN(entry->addr + entry->size);
751
752 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
753 register_nosave_region(PFN_UP(entry->addr), pfn);
754
755 if (pfn >= limit_pfn)
756 break;
757 }
758}
759
760#ifdef CONFIG_ACPI
761/*
762 * Register ACPI NVS memory regions, so that we can save/restore them during
763 * hibernation and the subsequent resume:
764 */
765static int __init e820__register_nvs_regions(void)
766{
767 int i;
768
769 for (i = 0; i < e820_table->nr_entries; i++) {
770 struct e820_entry *entry = &e820_table->entries[i];
771
772 if (entry->type == E820_TYPE_NVS)
773 acpi_nvs_register(entry->addr, entry->size);
774 }
775
776 return 0;
777}
778core_initcall(e820__register_nvs_regions);
779#endif
780
781/*
782 * Allocate the requested number of bytes with the requsted alignment
783 * and return (the physical address) to the caller. Also register this
784 * range in the 'kexec' E820 table as a reserved range.
785 *
786 * This allows kexec to fake a new mptable, as if it came from the real
787 * system.
788 */
789u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
790{
791 u64 addr;
792
793 addr = memblock_phys_alloc(size, align);
794 if (addr) {
795 e820__range_update_kexec(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
796 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
797 e820__update_table_kexec();
798 }
799
800 return addr;
801}
802
803#ifdef CONFIG_X86_32
804# ifdef CONFIG_X86_PAE
805# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
806# else
807# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
808# endif
809#else /* CONFIG_X86_32 */
810# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
811#endif
812
813/*
814 * Find the highest page frame number we have available
815 */
816static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
817{
818 int i;
819 unsigned long last_pfn = 0;
820 unsigned long max_arch_pfn = MAX_ARCH_PFN;
821
822 for (i = 0; i < e820_table->nr_entries; i++) {
823 struct e820_entry *entry = &e820_table->entries[i];
824 unsigned long start_pfn;
825 unsigned long end_pfn;
826
827 if (entry->type != type)
828 continue;
829
830 start_pfn = entry->addr >> PAGE_SHIFT;
831 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
832
833 if (start_pfn >= limit_pfn)
834 continue;
835 if (end_pfn > limit_pfn) {
836 last_pfn = limit_pfn;
837 break;
838 }
839 if (end_pfn > last_pfn)
840 last_pfn = end_pfn;
841 }
842
843 if (last_pfn > max_arch_pfn)
844 last_pfn = max_arch_pfn;
845
846 pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
847 last_pfn, max_arch_pfn);
848 return last_pfn;
849}
850
851unsigned long __init e820__end_of_ram_pfn(void)
852{
853 return e820_end_pfn(MAX_ARCH_PFN, E820_TYPE_RAM);
854}
855
856unsigned long __init e820__end_of_low_ram_pfn(void)
857{
858 return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
859}
860
861static void __init early_panic(char *msg)
862{
863 early_printk(msg);
864 panic(msg);
865}
866
867static int userdef __initdata;
868
869/* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
870static int __init parse_memopt(char *p)
871{
872 u64 mem_size;
873
874 if (!p)
875 return -EINVAL;
876
877 if (!strcmp(p, "nopentium")) {
878#ifdef CONFIG_X86_32
879 setup_clear_cpu_cap(X86_FEATURE_PSE);
880 return 0;
881#else
882 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
883 return -EINVAL;
884#endif
885 }
886
887 userdef = 1;
888 mem_size = memparse(p, &p);
889
890 /* Don't remove all memory when getting "mem={invalid}" parameter: */
891 if (mem_size == 0)
892 return -EINVAL;
893
894 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
895
896#ifdef CONFIG_MEMORY_HOTPLUG
897 max_mem_size = mem_size;
898#endif
899
900 return 0;
901}
902early_param("mem", parse_memopt);
903
904static int __init parse_memmap_one(char *p)
905{
906 char *oldp;
907 u64 start_at, mem_size;
908
909 if (!p)
910 return -EINVAL;
911
912 if (!strncmp(p, "exactmap", 8)) {
913 e820_table->nr_entries = 0;
914 userdef = 1;
915 return 0;
916 }
917
918 oldp = p;
919 mem_size = memparse(p, &p);
920 if (p == oldp)
921 return -EINVAL;
922
923 userdef = 1;
924 if (*p == '@') {
925 start_at = memparse(p+1, &p);
926 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
927 } else if (*p == '#') {
928 start_at = memparse(p+1, &p);
929 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
930 } else if (*p == '$') {
931 start_at = memparse(p+1, &p);
932 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
933 } else if (*p == '!') {
934 start_at = memparse(p+1, &p);
935 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
936 } else if (*p == '%') {
937 enum e820_type from = 0, to = 0;
938
939 start_at = memparse(p + 1, &p);
940 if (*p == '-')
941 from = simple_strtoull(p + 1, &p, 0);
942 if (*p == '+')
943 to = simple_strtoull(p + 1, &p, 0);
944 if (*p != '\0')
945 return -EINVAL;
946 if (from && to)
947 e820__range_update(start_at, mem_size, from, to);
948 else if (to)
949 e820__range_add(start_at, mem_size, to);
950 else if (from)
951 e820__range_remove(start_at, mem_size, from, 1);
952 else
953 e820__range_remove(start_at, mem_size, 0, 0);
954 } else {
955 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
956 }
957
958 return *p == '\0' ? 0 : -EINVAL;
959}
960
961static int __init parse_memmap_opt(char *str)
962{
963 while (str) {
964 char *k = strchr(str, ',');
965
966 if (k)
967 *k++ = 0;
968
969 parse_memmap_one(str);
970 str = k;
971 }
972
973 return 0;
974}
975early_param("memmap", parse_memmap_opt);
976
977/*
978 * Reserve all entries from the bootloader's extensible data nodes list,
979 * because if present we are going to use it later on to fetch e820
980 * entries from it:
981 */
982void __init e820__reserve_setup_data(void)
983{
984 struct setup_data *data;
985 u64 pa_data;
986
987 pa_data = boot_params.hdr.setup_data;
988 if (!pa_data)
989 return;
990
991 while (pa_data) {
992 data = early_memremap(pa_data, sizeof(*data));
993 e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
994
995 /*
996 * SETUP_EFI is supplied by kexec and does not need to be
997 * reserved.
998 */
999 if (data->type != SETUP_EFI)
1000 e820__range_update_kexec(pa_data,
1001 sizeof(*data) + data->len,
1002 E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1003
1004 if (data->type == SETUP_INDIRECT &&
1005 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT) {
1006 e820__range_update(((struct setup_indirect *)data->data)->addr,
1007 ((struct setup_indirect *)data->data)->len,
1008 E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1009 e820__range_update_kexec(((struct setup_indirect *)data->data)->addr,
1010 ((struct setup_indirect *)data->data)->len,
1011 E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1012 }
1013
1014 pa_data = data->next;
1015 early_memunmap(data, sizeof(*data));
1016 }
1017
1018 e820__update_table(e820_table);
1019 e820__update_table(e820_table_kexec);
1020
1021 pr_info("extended physical RAM map:\n");
1022 e820__print_table("reserve setup_data");
1023}
1024
1025/*
1026 * Called after parse_early_param(), after early parameters (such as mem=)
1027 * have been processed, in which case we already have an E820 table filled in
1028 * via the parameter callback function(s), but it's not sorted and printed yet:
1029 */
1030void __init e820__finish_early_params(void)
1031{
1032 if (userdef) {
1033 if (e820__update_table(e820_table) < 0)
1034 early_panic("Invalid user supplied memory map");
1035
1036 pr_info("user-defined physical RAM map:\n");
1037 e820__print_table("user");
1038 }
1039}
1040
1041static const char *__init e820_type_to_string(struct e820_entry *entry)
1042{
1043 switch (entry->type) {
1044 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1045 case E820_TYPE_RAM: return "System RAM";
1046 case E820_TYPE_ACPI: return "ACPI Tables";
1047 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1048 case E820_TYPE_UNUSABLE: return "Unusable memory";
1049 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1050 case E820_TYPE_PMEM: return "Persistent Memory";
1051 case E820_TYPE_RESERVED: return "Reserved";
1052 case E820_TYPE_SOFT_RESERVED: return "Soft Reserved";
1053 default: return "Unknown E820 type";
1054 }
1055}
1056
1057static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1058{
1059 switch (entry->type) {
1060 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1061 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1062 case E820_TYPE_ACPI: /* Fall-through: */
1063 case E820_TYPE_NVS: /* Fall-through: */
1064 case E820_TYPE_UNUSABLE: /* Fall-through: */
1065 case E820_TYPE_PRAM: /* Fall-through: */
1066 case E820_TYPE_PMEM: /* Fall-through: */
1067 case E820_TYPE_RESERVED: /* Fall-through: */
1068 case E820_TYPE_SOFT_RESERVED: /* Fall-through: */
1069 default: return IORESOURCE_MEM;
1070 }
1071}
1072
1073static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1074{
1075 switch (entry->type) {
1076 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1077 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1078 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1079 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1080 case E820_TYPE_RESERVED: return IORES_DESC_RESERVED;
1081 case E820_TYPE_SOFT_RESERVED: return IORES_DESC_SOFT_RESERVED;
1082 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1083 case E820_TYPE_RAM: /* Fall-through: */
1084 case E820_TYPE_UNUSABLE: /* Fall-through: */
1085 default: return IORES_DESC_NONE;
1086 }
1087}
1088
1089static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1090{
1091 /* this is the legacy bios/dos rom-shadow + mmio region */
1092 if (res->start < (1ULL<<20))
1093 return true;
1094
1095 /*
1096 * Treat persistent memory and other special memory ranges like
1097 * device memory, i.e. reserve it for exclusive use of a driver
1098 */
1099 switch (type) {
1100 case E820_TYPE_RESERVED:
1101 case E820_TYPE_SOFT_RESERVED:
1102 case E820_TYPE_PRAM:
1103 case E820_TYPE_PMEM:
1104 return false;
1105 case E820_TYPE_RESERVED_KERN:
1106 case E820_TYPE_RAM:
1107 case E820_TYPE_ACPI:
1108 case E820_TYPE_NVS:
1109 case E820_TYPE_UNUSABLE:
1110 default:
1111 return true;
1112 }
1113}
1114
1115/*
1116 * Mark E820 reserved areas as busy for the resource manager:
1117 */
1118
1119static struct resource __initdata *e820_res;
1120
1121void __init e820__reserve_resources(void)
1122{
1123 int i;
1124 struct resource *res;
1125 u64 end;
1126
1127 res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1128 SMP_CACHE_BYTES);
1129 if (!res)
1130 panic("%s: Failed to allocate %zu bytes\n", __func__,
1131 sizeof(*res) * e820_table->nr_entries);
1132 e820_res = res;
1133
1134 for (i = 0; i < e820_table->nr_entries; i++) {
1135 struct e820_entry *entry = e820_table->entries + i;
1136
1137 end = entry->addr + entry->size - 1;
1138 if (end != (resource_size_t)end) {
1139 res++;
1140 continue;
1141 }
1142 res->start = entry->addr;
1143 res->end = end;
1144 res->name = e820_type_to_string(entry);
1145 res->flags = e820_type_to_iomem_type(entry);
1146 res->desc = e820_type_to_iores_desc(entry);
1147
1148 /*
1149 * Don't register the region that could be conflicted with
1150 * PCI device BAR resources and insert them later in
1151 * pcibios_resource_survey():
1152 */
1153 if (do_mark_busy(entry->type, res)) {
1154 res->flags |= IORESOURCE_BUSY;
1155 insert_resource(&iomem_resource, res);
1156 }
1157 res++;
1158 }
1159
1160 /* Expose the bootloader-provided memory layout to the sysfs. */
1161 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1162 struct e820_entry *entry = e820_table_firmware->entries + i;
1163
1164 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1165 }
1166}
1167
1168/*
1169 * How much should we pad the end of RAM, depending on where it is?
1170 */
1171static unsigned long __init ram_alignment(resource_size_t pos)
1172{
1173 unsigned long mb = pos >> 20;
1174
1175 /* To 64kB in the first megabyte */
1176 if (!mb)
1177 return 64*1024;
1178
1179 /* To 1MB in the first 16MB */
1180 if (mb < 16)
1181 return 1024*1024;
1182
1183 /* To 64MB for anything above that */
1184 return 64*1024*1024;
1185}
1186
1187#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1188
1189void __init e820__reserve_resources_late(void)
1190{
1191 int i;
1192 struct resource *res;
1193
1194 res = e820_res;
1195 for (i = 0; i < e820_table->nr_entries; i++) {
1196 if (!res->parent && res->end)
1197 insert_resource_expand_to_fit(&iomem_resource, res);
1198 res++;
1199 }
1200
1201 /*
1202 * Try to bump up RAM regions to reasonable boundaries, to
1203 * avoid stolen RAM:
1204 */
1205 for (i = 0; i < e820_table->nr_entries; i++) {
1206 struct e820_entry *entry = &e820_table->entries[i];
1207 u64 start, end;
1208
1209 if (entry->type != E820_TYPE_RAM)
1210 continue;
1211
1212 start = entry->addr + entry->size;
1213 end = round_up(start, ram_alignment(start)) - 1;
1214 if (end > MAX_RESOURCE_SIZE)
1215 end = MAX_RESOURCE_SIZE;
1216 if (start >= end)
1217 continue;
1218
1219 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1220 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1221 }
1222}
1223
1224/*
1225 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1226 */
1227char *__init e820__memory_setup_default(void)
1228{
1229 char *who = "BIOS-e820";
1230
1231 /*
1232 * Try to copy the BIOS-supplied E820-map.
1233 *
1234 * Otherwise fake a memory map; one section from 0k->640k,
1235 * the next section from 1mb->appropriate_mem_k
1236 */
1237 if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1238 u64 mem_size;
1239
1240 /* Compare results from other methods and take the one that gives more RAM: */
1241 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1242 mem_size = boot_params.screen_info.ext_mem_k;
1243 who = "BIOS-88";
1244 } else {
1245 mem_size = boot_params.alt_mem_k;
1246 who = "BIOS-e801";
1247 }
1248
1249 e820_table->nr_entries = 0;
1250 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1251 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1252 }
1253
1254 /* We just appended a lot of ranges, sanitize the table: */
1255 e820__update_table(e820_table);
1256
1257 return who;
1258}
1259
1260/*
1261 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1262 * E820 map - with an optional platform quirk available for virtual platforms
1263 * to override this method of boot environment processing:
1264 */
1265void __init e820__memory_setup(void)
1266{
1267 char *who;
1268
1269 /* This is a firmware interface ABI - make sure we don't break it: */
1270 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1271
1272 who = x86_init.resources.memory_setup();
1273
1274 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1275 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1276
1277 pr_info("BIOS-provided physical RAM map:\n");
1278 e820__print_table(who);
1279}
1280
1281void __init e820__memblock_setup(void)
1282{
1283 int i;
1284 u64 end;
1285
1286 /*
1287 * The bootstrap memblock region count maximum is 128 entries
1288 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1289 * than that - so allow memblock resizing.
1290 *
1291 * This is safe, because this call happens pretty late during x86 setup,
1292 * so we know about reserved memory regions already. (This is important
1293 * so that memblock resizing does no stomp over reserved areas.)
1294 */
1295 memblock_allow_resize();
1296
1297 for (i = 0; i < e820_table->nr_entries; i++) {
1298 struct e820_entry *entry = &e820_table->entries[i];
1299
1300 end = entry->addr + entry->size;
1301 if (end != (resource_size_t)end)
1302 continue;
1303
1304 if (entry->type == E820_TYPE_SOFT_RESERVED)
1305 memblock_reserve(entry->addr, entry->size);
1306
1307 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1308 continue;
1309
1310 memblock_add(entry->addr, entry->size);
1311 }
1312
1313 /* Throw away partial pages: */
1314 memblock_trim_memory(PAGE_SIZE);
1315
1316 memblock_dump_all();
1317}