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/export.h>
16 #include <linux/bootmem.h>
17 #include <linux/pfn.h>
18 #include <linux/suspend.h>
19 #include <linux/acpi.h>
20 #include <linux/firmware-map.h>
21 #include <linux/memblock.h>
22 #include <linux/sort.h>
23
24 #include <asm/e820.h>
25 #include <asm/proto.h>
26 #include <asm/setup.h>
27
28 /*
29 * The e820 map is the map that gets modified e.g. with command line parameters
30 * and that is also registered with modifications in the kernel resource tree
31 * with the iomem_resource as parent.
32 *
33 * The e820_saved is directly saved after the BIOS-provided memory map is
34 * copied. It doesn't get modified afterwards. It's registered for the
35 * /sys/firmware/memmap interface.
36 *
37 * That memory map is not modified and is used as base for kexec. The kexec'd
38 * kernel should get the same memory map as the firmware provides. Then the
39 * user can e.g. boot the original kernel with mem=1G while still booting the
40 * next kernel with full memory.
41 */
42 struct e820map e820;
43 struct e820map e820_saved;
44
45 /* For PCI or other memory-mapped resources */
46 unsigned long pci_mem_start = 0xaeedbabe;
47 #ifdef CONFIG_PCI
48 EXPORT_SYMBOL(pci_mem_start);
49 #endif
50
51 /*
52 * This function checks if any part of the range <start,end> is mapped
53 * with type.
54 */
55 int
e820_any_mapped(u64 start,u64 end,unsigned type)56 e820_any_mapped(u64 start, u64 end, unsigned type)
57 {
58 int i;
59
60 for (i = 0; i < e820.nr_map; i++) {
61 struct e820entry *ei = &e820.map[i];
62
63 if (type && ei->type != type)
64 continue;
65 if (ei->addr >= end || ei->addr + ei->size <= start)
66 continue;
67 return 1;
68 }
69 return 0;
70 }
71 EXPORT_SYMBOL_GPL(e820_any_mapped);
72
73 /*
74 * This function checks if the entire range <start,end> is mapped with type.
75 *
76 * Note: this function only works correct if the e820 table is sorted and
77 * not-overlapping, which is the case
78 */
e820_all_mapped(u64 start,u64 end,unsigned type)79 int __init e820_all_mapped(u64 start, u64 end, unsigned type)
80 {
81 int i;
82
83 for (i = 0; i < e820.nr_map; i++) {
84 struct e820entry *ei = &e820.map[i];
85
86 if (type && ei->type != type)
87 continue;
88 /* is the region (part) in overlap with the current region ?*/
89 if (ei->addr >= end || ei->addr + ei->size <= start)
90 continue;
91
92 /* if the region is at the beginning of <start,end> we move
93 * start to the end of the region since it's ok until there
94 */
95 if (ei->addr <= start)
96 start = ei->addr + ei->size;
97 /*
98 * if start is now at or beyond end, we're done, full
99 * coverage
100 */
101 if (start >= end)
102 return 1;
103 }
104 return 0;
105 }
106
107 /*
108 * Add a memory region to the kernel e820 map.
109 */
__e820_add_region(struct e820map * e820x,u64 start,u64 size,int type)110 static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
111 int type)
112 {
113 int x = e820x->nr_map;
114
115 if (x >= ARRAY_SIZE(e820x->map)) {
116 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
117 return;
118 }
119
120 e820x->map[x].addr = start;
121 e820x->map[x].size = size;
122 e820x->map[x].type = type;
123 e820x->nr_map++;
124 }
125
e820_add_region(u64 start,u64 size,int type)126 void __init e820_add_region(u64 start, u64 size, int type)
127 {
128 __e820_add_region(&e820, start, size, type);
129 }
130
e820_print_type(u32 type)131 static void __init e820_print_type(u32 type)
132 {
133 switch (type) {
134 case E820_RAM:
135 case E820_RESERVED_KERN:
136 printk(KERN_CONT "(usable)");
137 break;
138 case E820_RESERVED:
139 printk(KERN_CONT "(reserved)");
140 break;
141 case E820_ACPI:
142 printk(KERN_CONT "(ACPI data)");
143 break;
144 case E820_NVS:
145 printk(KERN_CONT "(ACPI NVS)");
146 break;
147 case E820_UNUSABLE:
148 printk(KERN_CONT "(unusable)");
149 break;
150 default:
151 printk(KERN_CONT "type %u", type);
152 break;
153 }
154 }
155
e820_print_map(char * who)156 void __init e820_print_map(char *who)
157 {
158 int i;
159
160 for (i = 0; i < e820.nr_map; i++) {
161 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
162 (unsigned long long) e820.map[i].addr,
163 (unsigned long long)
164 (e820.map[i].addr + e820.map[i].size));
165 e820_print_type(e820.map[i].type);
166 printk(KERN_CONT "\n");
167 }
168 }
169
170 /*
171 * Sanitize the BIOS e820 map.
172 *
173 * Some e820 responses include overlapping entries. The following
174 * replaces the original e820 map with a new one, removing overlaps,
175 * and resolving conflicting memory types in favor of highest
176 * numbered type.
177 *
178 * The input parameter biosmap points to an array of 'struct
179 * e820entry' which on entry has elements in the range [0, *pnr_map)
180 * valid, and which has space for up to max_nr_map entries.
181 * On return, the resulting sanitized e820 map entries will be in
182 * overwritten in the same location, starting at biosmap.
183 *
184 * The integer pointed to by pnr_map must be valid on entry (the
185 * current number of valid entries located at biosmap) and will
186 * be updated on return, with the new number of valid entries
187 * (something no more than max_nr_map.)
188 *
189 * The return value from sanitize_e820_map() is zero if it
190 * successfully 'sanitized' the map entries passed in, and is -1
191 * if it did nothing, which can happen if either of (1) it was
192 * only passed one map entry, or (2) any of the input map entries
193 * were invalid (start + size < start, meaning that the size was
194 * so big the described memory range wrapped around through zero.)
195 *
196 * Visually we're performing the following
197 * (1,2,3,4 = memory types)...
198 *
199 * Sample memory map (w/overlaps):
200 * ____22__________________
201 * ______________________4_
202 * ____1111________________
203 * _44_____________________
204 * 11111111________________
205 * ____________________33__
206 * ___________44___________
207 * __________33333_________
208 * ______________22________
209 * ___________________2222_
210 * _________111111111______
211 * _____________________11_
212 * _________________4______
213 *
214 * Sanitized equivalent (no overlap):
215 * 1_______________________
216 * _44_____________________
217 * ___1____________________
218 * ____22__________________
219 * ______11________________
220 * _________1______________
221 * __________3_____________
222 * ___________44___________
223 * _____________33_________
224 * _______________2________
225 * ________________1_______
226 * _________________4______
227 * ___________________2____
228 * ____________________33__
229 * ______________________4_
230 */
231 struct change_member {
232 struct e820entry *pbios; /* pointer to original bios entry */
233 unsigned long long addr; /* address for this change point */
234 };
235
cpcompare(const void * a,const void * b)236 static int __init cpcompare(const void *a, const void *b)
237 {
238 struct change_member * const *app = a, * const *bpp = b;
239 const struct change_member *ap = *app, *bp = *bpp;
240
241 /*
242 * Inputs are pointers to two elements of change_point[]. If their
243 * addresses are unequal, their difference dominates. If the addresses
244 * are equal, then consider one that represents the end of its region
245 * to be greater than one that does not.
246 */
247 if (ap->addr != bp->addr)
248 return ap->addr > bp->addr ? 1 : -1;
249
250 return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr);
251 }
252
sanitize_e820_map(struct e820entry * biosmap,int max_nr_map,u32 * pnr_map)253 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
254 u32 *pnr_map)
255 {
256 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
257 static struct change_member *change_point[2*E820_X_MAX] __initdata;
258 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
259 static struct e820entry new_bios[E820_X_MAX] __initdata;
260 unsigned long current_type, last_type;
261 unsigned long long last_addr;
262 int chgidx;
263 int overlap_entries;
264 int new_bios_entry;
265 int old_nr, new_nr, chg_nr;
266 int i;
267
268 /* if there's only one memory region, don't bother */
269 if (*pnr_map < 2)
270 return -1;
271
272 old_nr = *pnr_map;
273 BUG_ON(old_nr > max_nr_map);
274
275 /* bail out if we find any unreasonable addresses in bios map */
276 for (i = 0; i < old_nr; i++)
277 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
278 return -1;
279
280 /* create pointers for initial change-point information (for sorting) */
281 for (i = 0; i < 2 * old_nr; i++)
282 change_point[i] = &change_point_list[i];
283
284 /* record all known change-points (starting and ending addresses),
285 omitting those that are for empty memory regions */
286 chgidx = 0;
287 for (i = 0; i < old_nr; i++) {
288 if (biosmap[i].size != 0) {
289 change_point[chgidx]->addr = biosmap[i].addr;
290 change_point[chgidx++]->pbios = &biosmap[i];
291 change_point[chgidx]->addr = biosmap[i].addr +
292 biosmap[i].size;
293 change_point[chgidx++]->pbios = &biosmap[i];
294 }
295 }
296 chg_nr = chgidx;
297
298 /* sort change-point list by memory addresses (low -> high) */
299 sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL);
300
301 /* create a new bios memory map, removing overlaps */
302 overlap_entries = 0; /* number of entries in the overlap table */
303 new_bios_entry = 0; /* index for creating new bios map entries */
304 last_type = 0; /* start with undefined memory type */
305 last_addr = 0; /* start with 0 as last starting address */
306
307 /* loop through change-points, determining affect on the new bios map */
308 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
309 /* keep track of all overlapping bios entries */
310 if (change_point[chgidx]->addr ==
311 change_point[chgidx]->pbios->addr) {
312 /*
313 * add map entry to overlap list (> 1 entry
314 * implies an overlap)
315 */
316 overlap_list[overlap_entries++] =
317 change_point[chgidx]->pbios;
318 } else {
319 /*
320 * remove entry from list (order independent,
321 * so swap with last)
322 */
323 for (i = 0; i < overlap_entries; i++) {
324 if (overlap_list[i] ==
325 change_point[chgidx]->pbios)
326 overlap_list[i] =
327 overlap_list[overlap_entries-1];
328 }
329 overlap_entries--;
330 }
331 /*
332 * if there are overlapping entries, decide which
333 * "type" to use (larger value takes precedence --
334 * 1=usable, 2,3,4,4+=unusable)
335 */
336 current_type = 0;
337 for (i = 0; i < overlap_entries; i++)
338 if (overlap_list[i]->type > current_type)
339 current_type = overlap_list[i]->type;
340 /*
341 * continue building up new bios map based on this
342 * information
343 */
344 if (current_type != last_type) {
345 if (last_type != 0) {
346 new_bios[new_bios_entry].size =
347 change_point[chgidx]->addr - last_addr;
348 /*
349 * move forward only if the new size
350 * was non-zero
351 */
352 if (new_bios[new_bios_entry].size != 0)
353 /*
354 * no more space left for new
355 * bios entries ?
356 */
357 if (++new_bios_entry >= max_nr_map)
358 break;
359 }
360 if (current_type != 0) {
361 new_bios[new_bios_entry].addr =
362 change_point[chgidx]->addr;
363 new_bios[new_bios_entry].type = current_type;
364 last_addr = change_point[chgidx]->addr;
365 }
366 last_type = current_type;
367 }
368 }
369 /* retain count for new bios entries */
370 new_nr = new_bios_entry;
371
372 /* copy new bios mapping into original location */
373 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
374 *pnr_map = new_nr;
375
376 return 0;
377 }
378
__append_e820_map(struct e820entry * biosmap,int nr_map)379 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
380 {
381 while (nr_map) {
382 u64 start = biosmap->addr;
383 u64 size = biosmap->size;
384 u64 end = start + size;
385 u32 type = biosmap->type;
386
387 /* Overflow in 64 bits? Ignore the memory map. */
388 if (start > end)
389 return -1;
390
391 e820_add_region(start, size, type);
392
393 biosmap++;
394 nr_map--;
395 }
396 return 0;
397 }
398
399 /*
400 * Copy the BIOS e820 map into a safe place.
401 *
402 * Sanity-check it while we're at it..
403 *
404 * If we're lucky and live on a modern system, the setup code
405 * will have given us a memory map that we can use to properly
406 * set up memory. If we aren't, we'll fake a memory map.
407 */
append_e820_map(struct e820entry * biosmap,int nr_map)408 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
409 {
410 /* Only one memory region (or negative)? Ignore it */
411 if (nr_map < 2)
412 return -1;
413
414 return __append_e820_map(biosmap, nr_map);
415 }
416
__e820_update_range(struct e820map * e820x,u64 start,u64 size,unsigned old_type,unsigned new_type)417 static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
418 u64 size, unsigned old_type,
419 unsigned new_type)
420 {
421 u64 end;
422 unsigned int i;
423 u64 real_updated_size = 0;
424
425 BUG_ON(old_type == new_type);
426
427 if (size > (ULLONG_MAX - start))
428 size = ULLONG_MAX - start;
429
430 end = start + size;
431 printk(KERN_DEBUG "e820 update range: %016Lx - %016Lx ",
432 (unsigned long long) start,
433 (unsigned long long) end);
434 e820_print_type(old_type);
435 printk(KERN_CONT " ==> ");
436 e820_print_type(new_type);
437 printk(KERN_CONT "\n");
438
439 for (i = 0; i < e820x->nr_map; i++) {
440 struct e820entry *ei = &e820x->map[i];
441 u64 final_start, final_end;
442 u64 ei_end;
443
444 if (ei->type != old_type)
445 continue;
446
447 ei_end = ei->addr + ei->size;
448 /* totally covered by new range? */
449 if (ei->addr >= start && ei_end <= end) {
450 ei->type = new_type;
451 real_updated_size += ei->size;
452 continue;
453 }
454
455 /* new range is totally covered? */
456 if (ei->addr < start && ei_end > end) {
457 __e820_add_region(e820x, start, size, new_type);
458 __e820_add_region(e820x, end, ei_end - end, ei->type);
459 ei->size = start - ei->addr;
460 real_updated_size += size;
461 continue;
462 }
463
464 /* partially covered */
465 final_start = max(start, ei->addr);
466 final_end = min(end, ei_end);
467 if (final_start >= final_end)
468 continue;
469
470 __e820_add_region(e820x, final_start, final_end - final_start,
471 new_type);
472
473 real_updated_size += final_end - final_start;
474
475 /*
476 * left range could be head or tail, so need to update
477 * size at first.
478 */
479 ei->size -= final_end - final_start;
480 if (ei->addr < final_start)
481 continue;
482 ei->addr = final_end;
483 }
484 return real_updated_size;
485 }
486
e820_update_range(u64 start,u64 size,unsigned old_type,unsigned new_type)487 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
488 unsigned new_type)
489 {
490 return __e820_update_range(&e820, start, size, old_type, new_type);
491 }
492
e820_update_range_saved(u64 start,u64 size,unsigned old_type,unsigned new_type)493 static u64 __init e820_update_range_saved(u64 start, u64 size,
494 unsigned old_type, unsigned new_type)
495 {
496 return __e820_update_range(&e820_saved, start, size, old_type,
497 new_type);
498 }
499
500 /* make e820 not cover the range */
e820_remove_range(u64 start,u64 size,unsigned old_type,int checktype)501 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
502 int checktype)
503 {
504 int i;
505 u64 end;
506 u64 real_removed_size = 0;
507
508 if (size > (ULLONG_MAX - start))
509 size = ULLONG_MAX - start;
510
511 end = start + size;
512 printk(KERN_DEBUG "e820 remove range: %016Lx - %016Lx ",
513 (unsigned long long) start,
514 (unsigned long long) end);
515 if (checktype)
516 e820_print_type(old_type);
517 printk(KERN_CONT "\n");
518
519 for (i = 0; i < e820.nr_map; i++) {
520 struct e820entry *ei = &e820.map[i];
521 u64 final_start, final_end;
522 u64 ei_end;
523
524 if (checktype && ei->type != old_type)
525 continue;
526
527 ei_end = ei->addr + ei->size;
528 /* totally covered? */
529 if (ei->addr >= start && ei_end <= end) {
530 real_removed_size += ei->size;
531 memset(ei, 0, sizeof(struct e820entry));
532 continue;
533 }
534
535 /* new range is totally covered? */
536 if (ei->addr < start && ei_end > end) {
537 e820_add_region(end, ei_end - end, ei->type);
538 ei->size = start - ei->addr;
539 real_removed_size += size;
540 continue;
541 }
542
543 /* partially covered */
544 final_start = max(start, ei->addr);
545 final_end = min(end, ei_end);
546 if (final_start >= final_end)
547 continue;
548 real_removed_size += final_end - final_start;
549
550 /*
551 * left range could be head or tail, so need to update
552 * size at first.
553 */
554 ei->size -= final_end - final_start;
555 if (ei->addr < final_start)
556 continue;
557 ei->addr = final_end;
558 }
559 return real_removed_size;
560 }
561
update_e820(void)562 void __init update_e820(void)
563 {
564 u32 nr_map;
565
566 nr_map = e820.nr_map;
567 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
568 return;
569 e820.nr_map = nr_map;
570 printk(KERN_INFO "modified physical RAM map:\n");
571 e820_print_map("modified");
572 }
update_e820_saved(void)573 static void __init update_e820_saved(void)
574 {
575 u32 nr_map;
576
577 nr_map = e820_saved.nr_map;
578 if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
579 return;
580 e820_saved.nr_map = nr_map;
581 }
582 #define MAX_GAP_END 0x100000000ull
583 /*
584 * Search for a gap in the e820 memory space from start_addr to end_addr.
585 */
e820_search_gap(unsigned long * gapstart,unsigned long * gapsize,unsigned long start_addr,unsigned long long end_addr)586 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
587 unsigned long start_addr, unsigned long long end_addr)
588 {
589 unsigned long long last;
590 int i = e820.nr_map;
591 int found = 0;
592
593 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
594
595 while (--i >= 0) {
596 unsigned long long start = e820.map[i].addr;
597 unsigned long long end = start + e820.map[i].size;
598
599 if (end < start_addr)
600 continue;
601
602 /*
603 * Since "last" is at most 4GB, we know we'll
604 * fit in 32 bits if this condition is true
605 */
606 if (last > end) {
607 unsigned long gap = last - end;
608
609 if (gap >= *gapsize) {
610 *gapsize = gap;
611 *gapstart = end;
612 found = 1;
613 }
614 }
615 if (start < last)
616 last = start;
617 }
618 return found;
619 }
620
621 /*
622 * Search for the biggest gap in the low 32 bits of the e820
623 * memory space. We pass this space to PCI to assign MMIO resources
624 * for hotplug or unconfigured devices in.
625 * Hopefully the BIOS let enough space left.
626 */
e820_setup_gap(void)627 __init void e820_setup_gap(void)
628 {
629 unsigned long gapstart, gapsize;
630 int found;
631
632 gapstart = 0x10000000;
633 gapsize = 0x400000;
634 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
635
636 #ifdef CONFIG_X86_64
637 if (!found) {
638 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
639 printk(KERN_ERR
640 "PCI: Warning: Cannot find a gap in the 32bit address range\n"
641 "PCI: Unassigned devices with 32bit resource registers may break!\n");
642 }
643 #endif
644
645 /*
646 * e820_reserve_resources_late protect stolen RAM already
647 */
648 pci_mem_start = gapstart;
649
650 printk(KERN_INFO
651 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
652 pci_mem_start, gapstart, gapsize);
653 }
654
655 /**
656 * Because of the size limitation of struct boot_params, only first
657 * 128 E820 memory entries are passed to kernel via
658 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
659 * linked list of struct setup_data, which is parsed here.
660 */
parse_e820_ext(struct setup_data * sdata)661 void __init parse_e820_ext(struct setup_data *sdata)
662 {
663 int entries;
664 struct e820entry *extmap;
665
666 entries = sdata->len / sizeof(struct e820entry);
667 extmap = (struct e820entry *)(sdata->data);
668 __append_e820_map(extmap, entries);
669 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
670 printk(KERN_INFO "extended physical RAM map:\n");
671 e820_print_map("extended");
672 }
673
674 #if defined(CONFIG_X86_64) || \
675 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
676 /**
677 * Find the ranges of physical addresses that do not correspond to
678 * e820 RAM areas and mark the corresponding pages as nosave for
679 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
680 *
681 * This function requires the e820 map to be sorted and without any
682 * overlapping entries and assumes the first e820 area to be RAM.
683 */
e820_mark_nosave_regions(unsigned long limit_pfn)684 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
685 {
686 int i;
687 unsigned long pfn;
688
689 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
690 for (i = 1; i < e820.nr_map; i++) {
691 struct e820entry *ei = &e820.map[i];
692
693 if (pfn < PFN_UP(ei->addr))
694 register_nosave_region(pfn, PFN_UP(ei->addr));
695
696 pfn = PFN_DOWN(ei->addr + ei->size);
697 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
698 register_nosave_region(PFN_UP(ei->addr), pfn);
699
700 if (pfn >= limit_pfn)
701 break;
702 }
703 }
704 #endif
705
706 #ifdef CONFIG_ACPI
707 /**
708 * Mark ACPI NVS memory region, so that we can save/restore it during
709 * hibernation and the subsequent resume.
710 */
e820_mark_nvs_memory(void)711 static int __init e820_mark_nvs_memory(void)
712 {
713 int i;
714
715 for (i = 0; i < e820.nr_map; i++) {
716 struct e820entry *ei = &e820.map[i];
717
718 if (ei->type == E820_NVS)
719 acpi_nvs_register(ei->addr, ei->size);
720 }
721
722 return 0;
723 }
724 core_initcall(e820_mark_nvs_memory);
725 #endif
726
727 /*
728 * pre allocated 4k and reserved it in memblock and e820_saved
729 */
early_reserve_e820(u64 size,u64 align)730 u64 __init early_reserve_e820(u64 size, u64 align)
731 {
732 u64 addr;
733
734 addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
735 if (addr) {
736 e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED);
737 printk(KERN_INFO "update e820_saved for early_reserve_e820\n");
738 update_e820_saved();
739 }
740
741 return addr;
742 }
743
744 #ifdef CONFIG_X86_32
745 # ifdef CONFIG_X86_PAE
746 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
747 # else
748 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
749 # endif
750 #else /* CONFIG_X86_32 */
751 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
752 #endif
753
754 /*
755 * Find the highest page frame number we have available
756 */
e820_end_pfn(unsigned long limit_pfn,unsigned type)757 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
758 {
759 int i;
760 unsigned long last_pfn = 0;
761 unsigned long max_arch_pfn = MAX_ARCH_PFN;
762
763 for (i = 0; i < e820.nr_map; i++) {
764 struct e820entry *ei = &e820.map[i];
765 unsigned long start_pfn;
766 unsigned long end_pfn;
767
768 if (ei->type != type)
769 continue;
770
771 start_pfn = ei->addr >> PAGE_SHIFT;
772 end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
773
774 if (start_pfn >= limit_pfn)
775 continue;
776 if (end_pfn > limit_pfn) {
777 last_pfn = limit_pfn;
778 break;
779 }
780 if (end_pfn > last_pfn)
781 last_pfn = end_pfn;
782 }
783
784 if (last_pfn > max_arch_pfn)
785 last_pfn = max_arch_pfn;
786
787 printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
788 last_pfn, max_arch_pfn);
789 return last_pfn;
790 }
e820_end_of_ram_pfn(void)791 unsigned long __init e820_end_of_ram_pfn(void)
792 {
793 return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
794 }
795
e820_end_of_low_ram_pfn(void)796 unsigned long __init e820_end_of_low_ram_pfn(void)
797 {
798 return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
799 }
800
early_panic(char * msg)801 static void early_panic(char *msg)
802 {
803 early_printk(msg);
804 panic(msg);
805 }
806
807 static int userdef __initdata;
808
809 /* "mem=nopentium" disables the 4MB page tables. */
parse_memopt(char * p)810 static int __init parse_memopt(char *p)
811 {
812 u64 mem_size;
813
814 if (!p)
815 return -EINVAL;
816
817 if (!strcmp(p, "nopentium")) {
818 #ifdef CONFIG_X86_32
819 setup_clear_cpu_cap(X86_FEATURE_PSE);
820 return 0;
821 #else
822 printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
823 return -EINVAL;
824 #endif
825 }
826
827 userdef = 1;
828 mem_size = memparse(p, &p);
829 /* don't remove all of memory when handling "mem={invalid}" param */
830 if (mem_size == 0)
831 return -EINVAL;
832 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
833
834 return 0;
835 }
836 early_param("mem", parse_memopt);
837
parse_memmap_opt(char * p)838 static int __init parse_memmap_opt(char *p)
839 {
840 char *oldp;
841 u64 start_at, mem_size;
842
843 if (!p)
844 return -EINVAL;
845
846 if (!strncmp(p, "exactmap", 8)) {
847 #ifdef CONFIG_CRASH_DUMP
848 /*
849 * If we are doing a crash dump, we still need to know
850 * the real mem size before original memory map is
851 * reset.
852 */
853 saved_max_pfn = e820_end_of_ram_pfn();
854 #endif
855 e820.nr_map = 0;
856 userdef = 1;
857 return 0;
858 }
859
860 oldp = p;
861 mem_size = memparse(p, &p);
862 if (p == oldp)
863 return -EINVAL;
864
865 userdef = 1;
866 if (*p == '@') {
867 start_at = memparse(p+1, &p);
868 e820_add_region(start_at, mem_size, E820_RAM);
869 } else if (*p == '#') {
870 start_at = memparse(p+1, &p);
871 e820_add_region(start_at, mem_size, E820_ACPI);
872 } else if (*p == '$') {
873 start_at = memparse(p+1, &p);
874 e820_add_region(start_at, mem_size, E820_RESERVED);
875 } else
876 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
877
878 return *p == '\0' ? 0 : -EINVAL;
879 }
880 early_param("memmap", parse_memmap_opt);
881
finish_e820_parsing(void)882 void __init finish_e820_parsing(void)
883 {
884 if (userdef) {
885 u32 nr = e820.nr_map;
886
887 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
888 early_panic("Invalid user supplied memory map");
889 e820.nr_map = nr;
890
891 printk(KERN_INFO "user-defined physical RAM map:\n");
892 e820_print_map("user");
893 }
894 }
895
e820_type_to_string(int e820_type)896 static inline const char *e820_type_to_string(int e820_type)
897 {
898 switch (e820_type) {
899 case E820_RESERVED_KERN:
900 case E820_RAM: return "System RAM";
901 case E820_ACPI: return "ACPI Tables";
902 case E820_NVS: return "ACPI Non-volatile Storage";
903 case E820_UNUSABLE: return "Unusable memory";
904 default: return "reserved";
905 }
906 }
907
908 /*
909 * Mark e820 reserved areas as busy for the resource manager.
910 */
911 static struct resource __initdata *e820_res;
e820_reserve_resources(void)912 void __init e820_reserve_resources(void)
913 {
914 int i;
915 struct resource *res;
916 u64 end;
917
918 res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
919 e820_res = res;
920 for (i = 0; i < e820.nr_map; i++) {
921 end = e820.map[i].addr + e820.map[i].size - 1;
922 if (end != (resource_size_t)end) {
923 res++;
924 continue;
925 }
926 res->name = e820_type_to_string(e820.map[i].type);
927 res->start = e820.map[i].addr;
928 res->end = end;
929
930 res->flags = IORESOURCE_MEM;
931
932 /*
933 * don't register the region that could be conflicted with
934 * pci device BAR resource and insert them later in
935 * pcibios_resource_survey()
936 */
937 if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20)) {
938 res->flags |= IORESOURCE_BUSY;
939 insert_resource(&iomem_resource, res);
940 }
941 res++;
942 }
943
944 for (i = 0; i < e820_saved.nr_map; i++) {
945 struct e820entry *entry = &e820_saved.map[i];
946 firmware_map_add_early(entry->addr,
947 entry->addr + entry->size - 1,
948 e820_type_to_string(entry->type));
949 }
950 }
951
952 /* How much should we pad RAM ending depending on where it is? */
ram_alignment(resource_size_t pos)953 static unsigned long ram_alignment(resource_size_t pos)
954 {
955 unsigned long mb = pos >> 20;
956
957 /* To 64kB in the first megabyte */
958 if (!mb)
959 return 64*1024;
960
961 /* To 1MB in the first 16MB */
962 if (mb < 16)
963 return 1024*1024;
964
965 /* To 64MB for anything above that */
966 return 64*1024*1024;
967 }
968
969 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
970
e820_reserve_resources_late(void)971 void __init e820_reserve_resources_late(void)
972 {
973 int i;
974 struct resource *res;
975
976 res = e820_res;
977 for (i = 0; i < e820.nr_map; i++) {
978 if (!res->parent && res->end)
979 insert_resource_expand_to_fit(&iomem_resource, res);
980 res++;
981 }
982
983 /*
984 * Try to bump up RAM regions to reasonable boundaries to
985 * avoid stolen RAM:
986 */
987 for (i = 0; i < e820.nr_map; i++) {
988 struct e820entry *entry = &e820.map[i];
989 u64 start, end;
990
991 if (entry->type != E820_RAM)
992 continue;
993 start = entry->addr + entry->size;
994 end = round_up(start, ram_alignment(start)) - 1;
995 if (end > MAX_RESOURCE_SIZE)
996 end = MAX_RESOURCE_SIZE;
997 if (start >= end)
998 continue;
999 printk(KERN_DEBUG "reserve RAM buffer: %016llx - %016llx ",
1000 start, end);
1001 reserve_region_with_split(&iomem_resource, start, end,
1002 "RAM buffer");
1003 }
1004 }
1005
default_machine_specific_memory_setup(void)1006 char *__init default_machine_specific_memory_setup(void)
1007 {
1008 char *who = "BIOS-e820";
1009 u32 new_nr;
1010 /*
1011 * Try to copy the BIOS-supplied E820-map.
1012 *
1013 * Otherwise fake a memory map; one section from 0k->640k,
1014 * the next section from 1mb->appropriate_mem_k
1015 */
1016 new_nr = boot_params.e820_entries;
1017 sanitize_e820_map(boot_params.e820_map,
1018 ARRAY_SIZE(boot_params.e820_map),
1019 &new_nr);
1020 boot_params.e820_entries = new_nr;
1021 if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1022 < 0) {
1023 u64 mem_size;
1024
1025 /* compare results from other methods and take the greater */
1026 if (boot_params.alt_mem_k
1027 < boot_params.screen_info.ext_mem_k) {
1028 mem_size = boot_params.screen_info.ext_mem_k;
1029 who = "BIOS-88";
1030 } else {
1031 mem_size = boot_params.alt_mem_k;
1032 who = "BIOS-e801";
1033 }
1034
1035 e820.nr_map = 0;
1036 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1037 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1038 }
1039
1040 /* In case someone cares... */
1041 return who;
1042 }
1043
setup_memory_map(void)1044 void __init setup_memory_map(void)
1045 {
1046 char *who;
1047
1048 who = x86_init.resources.memory_setup();
1049 memcpy(&e820_saved, &e820, sizeof(struct e820map));
1050 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1051 e820_print_map(who);
1052 }
1053
memblock_x86_fill(void)1054 void __init memblock_x86_fill(void)
1055 {
1056 int i;
1057 u64 end;
1058
1059 /*
1060 * EFI may have more than 128 entries
1061 * We are safe to enable resizing, beause memblock_x86_fill()
1062 * is rather later for x86
1063 */
1064 memblock_allow_resize();
1065
1066 for (i = 0; i < e820.nr_map; i++) {
1067 struct e820entry *ei = &e820.map[i];
1068
1069 end = ei->addr + ei->size;
1070 if (end != (resource_size_t)end)
1071 continue;
1072
1073 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1074 continue;
1075
1076 memblock_add(ei->addr, ei->size);
1077 }
1078
1079 /* throw away partial pages */
1080 memblock_trim_memory(PAGE_SIZE);
1081
1082 memblock_dump_all();
1083 }
1084
memblock_find_dma_reserve(void)1085 void __init memblock_find_dma_reserve(void)
1086 {
1087 #ifdef CONFIG_X86_64
1088 u64 nr_pages = 0, nr_free_pages = 0;
1089 unsigned long start_pfn, end_pfn;
1090 phys_addr_t start, end;
1091 int i;
1092 u64 u;
1093
1094 /*
1095 * need to find out used area below MAX_DMA_PFN
1096 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1097 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1098 */
1099 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1100 start_pfn = min_t(unsigned long, start_pfn, MAX_DMA_PFN);
1101 end_pfn = min_t(unsigned long, end_pfn, MAX_DMA_PFN);
1102 nr_pages += end_pfn - start_pfn;
1103 }
1104
1105 for_each_free_mem_range(u, MAX_NUMNODES, &start, &end, NULL) {
1106 start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1107 end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1108 if (start_pfn < end_pfn)
1109 nr_free_pages += end_pfn - start_pfn;
1110 }
1111
1112 set_dma_reserve(nr_pages - nr_free_pages);
1113 #endif
1114 }
1115