aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel/e820.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/kernel/e820.c')
-rw-r--r--arch/x86/kernel/e820.c1390
1 files changed, 1390 insertions, 0 deletions
diff --git a/arch/x86/kernel/e820.c b/arch/x86/kernel/e820.c
new file mode 100644
index 000000000000..28c29180b380
--- /dev/null
+++ b/arch/x86/kernel/e820.c
@@ -0,0 +1,1390 @@
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/bootmem.h>
15#include <linux/ioport.h>
16#include <linux/string.h>
17#include <linux/kexec.h>
18#include <linux/module.h>
19#include <linux/mm.h>
20#include <linux/pfn.h>
21#include <linux/suspend.h>
22#include <linux/firmware-map.h>
23
24#include <asm/pgtable.h>
25#include <asm/page.h>
26#include <asm/e820.h>
27#include <asm/proto.h>
28#include <asm/setup.h>
29#include <asm/trampoline.h>
30
31/*
32 * The e820 map is the map that gets modified e.g. with command line parameters
33 * and that is also registered with modifications in the kernel resource tree
34 * with the iomem_resource as parent.
35 *
36 * The e820_saved is directly saved after the BIOS-provided memory map is
37 * copied. It doesn't get modified afterwards. It's registered for the
38 * /sys/firmware/memmap interface.
39 *
40 * That memory map is not modified and is used as base for kexec. The kexec'd
41 * kernel should get the same memory map as the firmware provides. Then the
42 * user can e.g. boot the original kernel with mem=1G while still booting the
43 * next kernel with full memory.
44 */
45struct e820map e820;
46struct e820map e820_saved;
47
48/* For PCI or other memory-mapped resources */
49unsigned long pci_mem_start = 0xaeedbabe;
50#ifdef CONFIG_PCI
51EXPORT_SYMBOL(pci_mem_start);
52#endif
53
54/*
55 * This function checks if any part of the range <start,end> is mapped
56 * with type.
57 */
58int
59e820_any_mapped(u64 start, u64 end, unsigned type)
60{
61 int i;
62
63 for (i = 0; i < e820.nr_map; i++) {
64 struct e820entry *ei = &e820.map[i];
65
66 if (type && ei->type != type)
67 continue;
68 if (ei->addr >= end || ei->addr + ei->size <= start)
69 continue;
70 return 1;
71 }
72 return 0;
73}
74EXPORT_SYMBOL_GPL(e820_any_mapped);
75
76/*
77 * This function checks if the entire range <start,end> is mapped with type.
78 *
79 * Note: this function only works correct if the e820 table is sorted and
80 * not-overlapping, which is the case
81 */
82int __init e820_all_mapped(u64 start, u64 end, unsigned type)
83{
84 int i;
85
86 for (i = 0; i < e820.nr_map; i++) {
87 struct e820entry *ei = &e820.map[i];
88
89 if (type && ei->type != type)
90 continue;
91 /* is the region (part) in overlap with the current region ?*/
92 if (ei->addr >= end || ei->addr + ei->size <= start)
93 continue;
94
95 /* if the region is at the beginning of <start,end> we move
96 * start to the end of the region since it's ok until there
97 */
98 if (ei->addr <= start)
99 start = ei->addr + ei->size;
100 /*
101 * if start is now at or beyond end, we're done, full
102 * coverage
103 */
104 if (start >= end)
105 return 1;
106 }
107 return 0;
108}
109
110/*
111 * Add a memory region to the kernel e820 map.
112 */
113void __init e820_add_region(u64 start, u64 size, int type)
114{
115 int x = e820.nr_map;
116
117 if (x == ARRAY_SIZE(e820.map)) {
118 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
119 return;
120 }
121
122 e820.map[x].addr = start;
123 e820.map[x].size = size;
124 e820.map[x].type = type;
125 e820.nr_map++;
126}
127
128void __init e820_print_map(char *who)
129{
130 int i;
131
132 for (i = 0; i < e820.nr_map; i++) {
133 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
134 (unsigned long long) e820.map[i].addr,
135 (unsigned long long)
136 (e820.map[i].addr + e820.map[i].size));
137 switch (e820.map[i].type) {
138 case E820_RAM:
139 case E820_RESERVED_KERN:
140 printk(KERN_CONT "(usable)\n");
141 break;
142 case E820_RESERVED:
143 printk(KERN_CONT "(reserved)\n");
144 break;
145 case E820_ACPI:
146 printk(KERN_CONT "(ACPI data)\n");
147 break;
148 case E820_NVS:
149 printk(KERN_CONT "(ACPI NVS)\n");
150 break;
151 default:
152 printk(KERN_CONT "type %u\n", e820.map[i].type);
153 break;
154 }
155 }
156}
157
158/*
159 * Sanitize the BIOS e820 map.
160 *
161 * Some e820 responses include overlapping entries. The following
162 * replaces the original e820 map with a new one, removing overlaps,
163 * and resolving conflicting memory types in favor of highest
164 * numbered type.
165 *
166 * The input parameter biosmap points to an array of 'struct
167 * e820entry' which on entry has elements in the range [0, *pnr_map)
168 * valid, and which has space for up to max_nr_map entries.
169 * On return, the resulting sanitized e820 map entries will be in
170 * overwritten in the same location, starting at biosmap.
171 *
172 * The integer pointed to by pnr_map must be valid on entry (the
173 * current number of valid entries located at biosmap) and will
174 * be updated on return, with the new number of valid entries
175 * (something no more than max_nr_map.)
176 *
177 * The return value from sanitize_e820_map() is zero if it
178 * successfully 'sanitized' the map entries passed in, and is -1
179 * if it did nothing, which can happen if either of (1) it was
180 * only passed one map entry, or (2) any of the input map entries
181 * were invalid (start + size < start, meaning that the size was
182 * so big the described memory range wrapped around through zero.)
183 *
184 * Visually we're performing the following
185 * (1,2,3,4 = memory types)...
186 *
187 * Sample memory map (w/overlaps):
188 * ____22__________________
189 * ______________________4_
190 * ____1111________________
191 * _44_____________________
192 * 11111111________________
193 * ____________________33__
194 * ___________44___________
195 * __________33333_________
196 * ______________22________
197 * ___________________2222_
198 * _________111111111______
199 * _____________________11_
200 * _________________4______
201 *
202 * Sanitized equivalent (no overlap):
203 * 1_______________________
204 * _44_____________________
205 * ___1____________________
206 * ____22__________________
207 * ______11________________
208 * _________1______________
209 * __________3_____________
210 * ___________44___________
211 * _____________33_________
212 * _______________2________
213 * ________________1_______
214 * _________________4______
215 * ___________________2____
216 * ____________________33__
217 * ______________________4_
218 */
219
220int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
221 int *pnr_map)
222{
223 struct change_member {
224 struct e820entry *pbios; /* pointer to original bios entry */
225 unsigned long long addr; /* address for this change point */
226 };
227 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
228 static struct change_member *change_point[2*E820_X_MAX] __initdata;
229 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
230 static struct e820entry new_bios[E820_X_MAX] __initdata;
231 struct change_member *change_tmp;
232 unsigned long current_type, last_type;
233 unsigned long long last_addr;
234 int chgidx, still_changing;
235 int overlap_entries;
236 int new_bios_entry;
237 int old_nr, new_nr, chg_nr;
238 int i;
239
240 /* if there's only one memory region, don't bother */
241 if (*pnr_map < 2)
242 return -1;
243
244 old_nr = *pnr_map;
245 BUG_ON(old_nr > max_nr_map);
246
247 /* bail out if we find any unreasonable addresses in bios map */
248 for (i = 0; i < old_nr; i++)
249 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
250 return -1;
251
252 /* create pointers for initial change-point information (for sorting) */
253 for (i = 0; i < 2 * old_nr; i++)
254 change_point[i] = &change_point_list[i];
255
256 /* record all known change-points (starting and ending addresses),
257 omitting those that are for empty memory regions */
258 chgidx = 0;
259 for (i = 0; i < old_nr; i++) {
260 if (biosmap[i].size != 0) {
261 change_point[chgidx]->addr = biosmap[i].addr;
262 change_point[chgidx++]->pbios = &biosmap[i];
263 change_point[chgidx]->addr = biosmap[i].addr +
264 biosmap[i].size;
265 change_point[chgidx++]->pbios = &biosmap[i];
266 }
267 }
268 chg_nr = chgidx;
269
270 /* sort change-point list by memory addresses (low -> high) */
271 still_changing = 1;
272 while (still_changing) {
273 still_changing = 0;
274 for (i = 1; i < chg_nr; i++) {
275 unsigned long long curaddr, lastaddr;
276 unsigned long long curpbaddr, lastpbaddr;
277
278 curaddr = change_point[i]->addr;
279 lastaddr = change_point[i - 1]->addr;
280 curpbaddr = change_point[i]->pbios->addr;
281 lastpbaddr = change_point[i - 1]->pbios->addr;
282
283 /*
284 * swap entries, when:
285 *
286 * curaddr > lastaddr or
287 * curaddr == lastaddr and curaddr == curpbaddr and
288 * lastaddr != lastpbaddr
289 */
290 if (curaddr < lastaddr ||
291 (curaddr == lastaddr && curaddr == curpbaddr &&
292 lastaddr != lastpbaddr)) {
293 change_tmp = change_point[i];
294 change_point[i] = change_point[i-1];
295 change_point[i-1] = change_tmp;
296 still_changing = 1;
297 }
298 }
299 }
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
379static 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 */
408static 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
417static u64 __init e820_update_range_map(struct e820map *e820x, u64 start,
418 u64 size, unsigned old_type,
419 unsigned new_type)
420{
421 int i;
422 u64 real_updated_size = 0;
423
424 BUG_ON(old_type == new_type);
425
426 if (size > (ULLONG_MAX - start))
427 size = ULLONG_MAX - start;
428
429 for (i = 0; i < e820.nr_map; i++) {
430 struct e820entry *ei = &e820x->map[i];
431 u64 final_start, final_end;
432 if (ei->type != old_type)
433 continue;
434 /* totally covered? */
435 if (ei->addr >= start &&
436 (ei->addr + ei->size) <= (start + size)) {
437 ei->type = new_type;
438 real_updated_size += ei->size;
439 continue;
440 }
441 /* partially covered */
442 final_start = max(start, ei->addr);
443 final_end = min(start + size, ei->addr + ei->size);
444 if (final_start >= final_end)
445 continue;
446 e820_add_region(final_start, final_end - final_start,
447 new_type);
448 real_updated_size += final_end - final_start;
449
450 ei->size -= final_end - final_start;
451 if (ei->addr < final_start)
452 continue;
453 ei->addr = final_end;
454 }
455 return real_updated_size;
456}
457
458u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
459 unsigned new_type)
460{
461 return e820_update_range_map(&e820, start, size, old_type, new_type);
462}
463
464static u64 __init e820_update_range_saved(u64 start, u64 size,
465 unsigned old_type, unsigned new_type)
466{
467 return e820_update_range_map(&e820_saved, start, size, old_type,
468 new_type);
469}
470
471/* make e820 not cover the range */
472u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
473 int checktype)
474{
475 int i;
476 u64 real_removed_size = 0;
477
478 if (size > (ULLONG_MAX - start))
479 size = ULLONG_MAX - start;
480
481 for (i = 0; i < e820.nr_map; i++) {
482 struct e820entry *ei = &e820.map[i];
483 u64 final_start, final_end;
484
485 if (checktype && ei->type != old_type)
486 continue;
487 /* totally covered? */
488 if (ei->addr >= start &&
489 (ei->addr + ei->size) <= (start + size)) {
490 real_removed_size += ei->size;
491 memset(ei, 0, sizeof(struct e820entry));
492 continue;
493 }
494 /* partially covered */
495 final_start = max(start, ei->addr);
496 final_end = min(start + size, ei->addr + ei->size);
497 if (final_start >= final_end)
498 continue;
499 real_removed_size += final_end - final_start;
500
501 ei->size -= final_end - final_start;
502 if (ei->addr < final_start)
503 continue;
504 ei->addr = final_end;
505 }
506 return real_removed_size;
507}
508
509void __init update_e820(void)
510{
511 int nr_map;
512
513 nr_map = e820.nr_map;
514 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
515 return;
516 e820.nr_map = nr_map;
517 printk(KERN_INFO "modified physical RAM map:\n");
518 e820_print_map("modified");
519}
520static void __init update_e820_saved(void)
521{
522 int nr_map;
523
524 nr_map = e820_saved.nr_map;
525 if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
526 return;
527 e820_saved.nr_map = nr_map;
528}
529#define MAX_GAP_END 0x100000000ull
530/*
531 * Search for a gap in the e820 memory space from start_addr to end_addr.
532 */
533__init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
534 unsigned long start_addr, unsigned long long end_addr)
535{
536 unsigned long long last;
537 int i = e820.nr_map;
538 int found = 0;
539
540 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
541
542 while (--i >= 0) {
543 unsigned long long start = e820.map[i].addr;
544 unsigned long long end = start + e820.map[i].size;
545
546 if (end < start_addr)
547 continue;
548
549 /*
550 * Since "last" is at most 4GB, we know we'll
551 * fit in 32 bits if this condition is true
552 */
553 if (last > end) {
554 unsigned long gap = last - end;
555
556 if (gap >= *gapsize) {
557 *gapsize = gap;
558 *gapstart = end;
559 found = 1;
560 }
561 }
562 if (start < last)
563 last = start;
564 }
565 return found;
566}
567
568/*
569 * Search for the biggest gap in the low 32 bits of the e820
570 * memory space. We pass this space to PCI to assign MMIO resources
571 * for hotplug or unconfigured devices in.
572 * Hopefully the BIOS let enough space left.
573 */
574__init void e820_setup_gap(void)
575{
576 unsigned long gapstart, gapsize, round;
577 int found;
578
579 gapstart = 0x10000000;
580 gapsize = 0x400000;
581 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
582
583#ifdef CONFIG_X86_64
584 if (!found) {
585 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
586 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
587 "address range\n"
588 KERN_ERR "PCI: Unassigned devices with 32bit resource "
589 "registers may break!\n");
590 }
591#endif
592
593 /*
594 * See how much we want to round up: start off with
595 * rounding to the next 1MB area.
596 */
597 round = 0x100000;
598 while ((gapsize >> 4) > round)
599 round += round;
600 /* Fun with two's complement */
601 pci_mem_start = (gapstart + round) & -round;
602
603 printk(KERN_INFO
604 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
605 pci_mem_start, gapstart, gapsize);
606}
607
608/**
609 * Because of the size limitation of struct boot_params, only first
610 * 128 E820 memory entries are passed to kernel via
611 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
612 * linked list of struct setup_data, which is parsed here.
613 */
614void __init parse_e820_ext(struct setup_data *sdata, unsigned long pa_data)
615{
616 u32 map_len;
617 int entries;
618 struct e820entry *extmap;
619
620 entries = sdata->len / sizeof(struct e820entry);
621 map_len = sdata->len + sizeof(struct setup_data);
622 if (map_len > PAGE_SIZE)
623 sdata = early_ioremap(pa_data, map_len);
624 extmap = (struct e820entry *)(sdata->data);
625 __append_e820_map(extmap, entries);
626 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
627 if (map_len > PAGE_SIZE)
628 early_iounmap(sdata, map_len);
629 printk(KERN_INFO "extended physical RAM map:\n");
630 e820_print_map("extended");
631}
632
633#if defined(CONFIG_X86_64) || \
634 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
635/**
636 * Find the ranges of physical addresses that do not correspond to
637 * e820 RAM areas and mark the corresponding pages as nosave for
638 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
639 *
640 * This function requires the e820 map to be sorted and without any
641 * overlapping entries and assumes the first e820 area to be RAM.
642 */
643void __init e820_mark_nosave_regions(unsigned long limit_pfn)
644{
645 int i;
646 unsigned long pfn;
647
648 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
649 for (i = 1; i < e820.nr_map; i++) {
650 struct e820entry *ei = &e820.map[i];
651
652 if (pfn < PFN_UP(ei->addr))
653 register_nosave_region(pfn, PFN_UP(ei->addr));
654
655 pfn = PFN_DOWN(ei->addr + ei->size);
656 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
657 register_nosave_region(PFN_UP(ei->addr), pfn);
658
659 if (pfn >= limit_pfn)
660 break;
661 }
662}
663#endif
664
665/*
666 * Early reserved memory areas.
667 */
668#define MAX_EARLY_RES 20
669
670struct early_res {
671 u64 start, end;
672 char name[16];
673 char overlap_ok;
674};
675static struct early_res early_res[MAX_EARLY_RES] __initdata = {
676 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
677#if defined(CONFIG_X86_64) && defined(CONFIG_X86_TRAMPOLINE)
678 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + 2 * PAGE_SIZE, "TRAMPOLINE" },
679#endif
680#if defined(CONFIG_X86_32) && defined(CONFIG_SMP)
681 /*
682 * But first pinch a few for the stack/trampoline stuff
683 * FIXME: Don't need the extra page at 4K, but need to fix
684 * trampoline before removing it. (see the GDT stuff)
685 */
686 { PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE" },
687 /*
688 * Has to be in very low memory so we can execute
689 * real-mode AP code.
690 */
691 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + PAGE_SIZE, "TRAMPOLINE" },
692#endif
693 {}
694};
695
696static int __init find_overlapped_early(u64 start, u64 end)
697{
698 int i;
699 struct early_res *r;
700
701 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
702 r = &early_res[i];
703 if (end > r->start && start < r->end)
704 break;
705 }
706
707 return i;
708}
709
710/*
711 * Drop the i-th range from the early reservation map,
712 * by copying any higher ranges down one over it, and
713 * clearing what had been the last slot.
714 */
715static void __init drop_range(int i)
716{
717 int j;
718
719 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
720 ;
721
722 memmove(&early_res[i], &early_res[i + 1],
723 (j - 1 - i) * sizeof(struct early_res));
724
725 early_res[j - 1].end = 0;
726}
727
728/*
729 * Split any existing ranges that:
730 * 1) are marked 'overlap_ok', and
731 * 2) overlap with the stated range [start, end)
732 * into whatever portion (if any) of the existing range is entirely
733 * below or entirely above the stated range. Drop the portion
734 * of the existing range that overlaps with the stated range,
735 * which will allow the caller of this routine to then add that
736 * stated range without conflicting with any existing range.
737 */
738static void __init drop_overlaps_that_are_ok(u64 start, u64 end)
739{
740 int i;
741 struct early_res *r;
742 u64 lower_start, lower_end;
743 u64 upper_start, upper_end;
744 char name[16];
745
746 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
747 r = &early_res[i];
748
749 /* Continue past non-overlapping ranges */
750 if (end <= r->start || start >= r->end)
751 continue;
752
753 /*
754 * Leave non-ok overlaps as is; let caller
755 * panic "Overlapping early reservations"
756 * when it hits this overlap.
757 */
758 if (!r->overlap_ok)
759 return;
760
761 /*
762 * We have an ok overlap. We will drop it from the early
763 * reservation map, and add back in any non-overlapping
764 * portions (lower or upper) as separate, overlap_ok,
765 * non-overlapping ranges.
766 */
767
768 /* 1. Note any non-overlapping (lower or upper) ranges. */
769 strncpy(name, r->name, sizeof(name) - 1);
770
771 lower_start = lower_end = 0;
772 upper_start = upper_end = 0;
773 if (r->start < start) {
774 lower_start = r->start;
775 lower_end = start;
776 }
777 if (r->end > end) {
778 upper_start = end;
779 upper_end = r->end;
780 }
781
782 /* 2. Drop the original ok overlapping range */
783 drop_range(i);
784
785 i--; /* resume for-loop on copied down entry */
786
787 /* 3. Add back in any non-overlapping ranges. */
788 if (lower_end)
789 reserve_early_overlap_ok(lower_start, lower_end, name);
790 if (upper_end)
791 reserve_early_overlap_ok(upper_start, upper_end, name);
792 }
793}
794
795static void __init __reserve_early(u64 start, u64 end, char *name,
796 int overlap_ok)
797{
798 int i;
799 struct early_res *r;
800
801 i = find_overlapped_early(start, end);
802 if (i >= MAX_EARLY_RES)
803 panic("Too many early reservations");
804 r = &early_res[i];
805 if (r->end)
806 panic("Overlapping early reservations "
807 "%llx-%llx %s to %llx-%llx %s\n",
808 start, end - 1, name?name:"", r->start,
809 r->end - 1, r->name);
810 r->start = start;
811 r->end = end;
812 r->overlap_ok = overlap_ok;
813 if (name)
814 strncpy(r->name, name, sizeof(r->name) - 1);
815}
816
817/*
818 * A few early reservtations come here.
819 *
820 * The 'overlap_ok' in the name of this routine does -not- mean it
821 * is ok for these reservations to overlap an earlier reservation.
822 * Rather it means that it is ok for subsequent reservations to
823 * overlap this one.
824 *
825 * Use this entry point to reserve early ranges when you are doing
826 * so out of "Paranoia", reserving perhaps more memory than you need,
827 * just in case, and don't mind a subsequent overlapping reservation
828 * that is known to be needed.
829 *
830 * The drop_overlaps_that_are_ok() call here isn't really needed.
831 * It would be needed if we had two colliding 'overlap_ok'
832 * reservations, so that the second such would not panic on the
833 * overlap with the first. We don't have any such as of this
834 * writing, but might as well tolerate such if it happens in
835 * the future.
836 */
837void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
838{
839 drop_overlaps_that_are_ok(start, end);
840 __reserve_early(start, end, name, 1);
841}
842
843/*
844 * Most early reservations come here.
845 *
846 * We first have drop_overlaps_that_are_ok() drop any pre-existing
847 * 'overlap_ok' ranges, so that we can then reserve this memory
848 * range without risk of panic'ing on an overlapping overlap_ok
849 * early reservation.
850 */
851void __init reserve_early(u64 start, u64 end, char *name)
852{
853 drop_overlaps_that_are_ok(start, end);
854 __reserve_early(start, end, name, 0);
855}
856
857void __init free_early(u64 start, u64 end)
858{
859 struct early_res *r;
860 int i;
861
862 i = find_overlapped_early(start, end);
863 r = &early_res[i];
864 if (i >= MAX_EARLY_RES || r->end != end || r->start != start)
865 panic("free_early on not reserved area: %llx-%llx!",
866 start, end - 1);
867
868 drop_range(i);
869}
870
871void __init early_res_to_bootmem(u64 start, u64 end)
872{
873 int i, count;
874 u64 final_start, final_end;
875
876 count = 0;
877 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++)
878 count++;
879
880 printk(KERN_INFO "(%d early reservations) ==> bootmem\n", count);
881 for (i = 0; i < count; i++) {
882 struct early_res *r = &early_res[i];
883 printk(KERN_INFO " #%d [%010llx - %010llx] %16s", i,
884 r->start, r->end, r->name);
885 final_start = max(start, r->start);
886 final_end = min(end, r->end);
887 if (final_start >= final_end) {
888 printk(KERN_CONT "\n");
889 continue;
890 }
891 printk(KERN_CONT " ==> [%010llx - %010llx]\n",
892 final_start, final_end);
893 reserve_bootmem_generic(final_start, final_end - final_start,
894 BOOTMEM_DEFAULT);
895 }
896}
897
898/* Check for already reserved areas */
899static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
900{
901 int i;
902 u64 addr = *addrp;
903 int changed = 0;
904 struct early_res *r;
905again:
906 i = find_overlapped_early(addr, addr + size);
907 r = &early_res[i];
908 if (i < MAX_EARLY_RES && r->end) {
909 *addrp = addr = round_up(r->end, align);
910 changed = 1;
911 goto again;
912 }
913 return changed;
914}
915
916/* Check for already reserved areas */
917static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
918{
919 int i;
920 u64 addr = *addrp, last;
921 u64 size = *sizep;
922 int changed = 0;
923again:
924 last = addr + size;
925 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
926 struct early_res *r = &early_res[i];
927 if (last > r->start && addr < r->start) {
928 size = r->start - addr;
929 changed = 1;
930 goto again;
931 }
932 if (last > r->end && addr < r->end) {
933 addr = round_up(r->end, align);
934 size = last - addr;
935 changed = 1;
936 goto again;
937 }
938 if (last <= r->end && addr >= r->start) {
939 (*sizep)++;
940 return 0;
941 }
942 }
943 if (changed) {
944 *addrp = addr;
945 *sizep = size;
946 }
947 return changed;
948}
949
950/*
951 * Find a free area with specified alignment in a specific range.
952 */
953u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
954{
955 int i;
956
957 for (i = 0; i < e820.nr_map; i++) {
958 struct e820entry *ei = &e820.map[i];
959 u64 addr, last;
960 u64 ei_last;
961
962 if (ei->type != E820_RAM)
963 continue;
964 addr = round_up(ei->addr, align);
965 ei_last = ei->addr + ei->size;
966 if (addr < start)
967 addr = round_up(start, align);
968 if (addr >= ei_last)
969 continue;
970 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
971 ;
972 last = addr + size;
973 if (last > ei_last)
974 continue;
975 if (last > end)
976 continue;
977 return addr;
978 }
979 return -1ULL;
980}
981
982/*
983 * Find next free range after *start
984 */
985u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
986{
987 int i;
988
989 for (i = 0; i < e820.nr_map; i++) {
990 struct e820entry *ei = &e820.map[i];
991 u64 addr, last;
992 u64 ei_last;
993
994 if (ei->type != E820_RAM)
995 continue;
996 addr = round_up(ei->addr, align);
997 ei_last = ei->addr + ei->size;
998 if (addr < start)
999 addr = round_up(start, align);
1000 if (addr >= ei_last)
1001 continue;
1002 *sizep = ei_last - addr;
1003 while (bad_addr_size(&addr, sizep, align) &&
1004 addr + *sizep <= ei_last)
1005 ;
1006 last = addr + *sizep;
1007 if (last > ei_last)
1008 continue;
1009 return addr;
1010 }
1011 return -1UL;
1012
1013}
1014
1015/*
1016 * pre allocated 4k and reserved it in e820
1017 */
1018u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
1019{
1020 u64 size = 0;
1021 u64 addr;
1022 u64 start;
1023
1024 start = startt;
1025 while (size < sizet)
1026 start = find_e820_area_size(start, &size, align);
1027
1028 if (size < sizet)
1029 return 0;
1030
1031 addr = round_down(start + size - sizet, align);
1032 e820_update_range(addr, sizet, E820_RAM, E820_RESERVED);
1033 e820_update_range_saved(addr, sizet, E820_RAM, E820_RESERVED);
1034 printk(KERN_INFO "update e820 for early_reserve_e820\n");
1035 update_e820();
1036 update_e820_saved();
1037
1038 return addr;
1039}
1040
1041#ifdef CONFIG_X86_32
1042# ifdef CONFIG_X86_PAE
1043# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
1044# else
1045# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
1046# endif
1047#else /* CONFIG_X86_32 */
1048# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
1049#endif
1050
1051/*
1052 * Find the highest page frame number we have available
1053 */
1054static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
1055{
1056 int i;
1057 unsigned long last_pfn = 0;
1058 unsigned long max_arch_pfn = MAX_ARCH_PFN;
1059
1060 for (i = 0; i < e820.nr_map; i++) {
1061 struct e820entry *ei = &e820.map[i];
1062 unsigned long start_pfn;
1063 unsigned long end_pfn;
1064
1065 if (ei->type != type)
1066 continue;
1067
1068 start_pfn = ei->addr >> PAGE_SHIFT;
1069 end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
1070
1071 if (start_pfn >= limit_pfn)
1072 continue;
1073 if (end_pfn > limit_pfn) {
1074 last_pfn = limit_pfn;
1075 break;
1076 }
1077 if (end_pfn > last_pfn)
1078 last_pfn = end_pfn;
1079 }
1080
1081 if (last_pfn > max_arch_pfn)
1082 last_pfn = max_arch_pfn;
1083
1084 printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
1085 last_pfn, max_arch_pfn);
1086 return last_pfn;
1087}
1088unsigned long __init e820_end_of_ram_pfn(void)
1089{
1090 return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
1091}
1092
1093unsigned long __init e820_end_of_low_ram_pfn(void)
1094{
1095 return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
1096}
1097/*
1098 * Finds an active region in the address range from start_pfn to last_pfn and
1099 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
1100 */
1101int __init e820_find_active_region(const struct e820entry *ei,
1102 unsigned long start_pfn,
1103 unsigned long last_pfn,
1104 unsigned long *ei_startpfn,
1105 unsigned long *ei_endpfn)
1106{
1107 u64 align = PAGE_SIZE;
1108
1109 *ei_startpfn = round_up(ei->addr, align) >> PAGE_SHIFT;
1110 *ei_endpfn = round_down(ei->addr + ei->size, align) >> PAGE_SHIFT;
1111
1112 /* Skip map entries smaller than a page */
1113 if (*ei_startpfn >= *ei_endpfn)
1114 return 0;
1115
1116 /* Skip if map is outside the node */
1117 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
1118 *ei_startpfn >= last_pfn)
1119 return 0;
1120
1121 /* Check for overlaps */
1122 if (*ei_startpfn < start_pfn)
1123 *ei_startpfn = start_pfn;
1124 if (*ei_endpfn > last_pfn)
1125 *ei_endpfn = last_pfn;
1126
1127 return 1;
1128}
1129
1130/* Walk the e820 map and register active regions within a node */
1131void __init e820_register_active_regions(int nid, unsigned long start_pfn,
1132 unsigned long last_pfn)
1133{
1134 unsigned long ei_startpfn;
1135 unsigned long ei_endpfn;
1136 int i;
1137
1138 for (i = 0; i < e820.nr_map; i++)
1139 if (e820_find_active_region(&e820.map[i],
1140 start_pfn, last_pfn,
1141 &ei_startpfn, &ei_endpfn))
1142 add_active_range(nid, ei_startpfn, ei_endpfn);
1143}
1144
1145/*
1146 * Find the hole size (in bytes) in the memory range.
1147 * @start: starting address of the memory range to scan
1148 * @end: ending address of the memory range to scan
1149 */
1150u64 __init e820_hole_size(u64 start, u64 end)
1151{
1152 unsigned long start_pfn = start >> PAGE_SHIFT;
1153 unsigned long last_pfn = end >> PAGE_SHIFT;
1154 unsigned long ei_startpfn, ei_endpfn, ram = 0;
1155 int i;
1156
1157 for (i = 0; i < e820.nr_map; i++) {
1158 if (e820_find_active_region(&e820.map[i],
1159 start_pfn, last_pfn,
1160 &ei_startpfn, &ei_endpfn))
1161 ram += ei_endpfn - ei_startpfn;
1162 }
1163 return end - start - ((u64)ram << PAGE_SHIFT);
1164}
1165
1166static void early_panic(char *msg)
1167{
1168 early_printk(msg);
1169 panic(msg);
1170}
1171
1172static int userdef __initdata;
1173
1174/* "mem=nopentium" disables the 4MB page tables. */
1175static int __init parse_memopt(char *p)
1176{
1177 u64 mem_size;
1178
1179 if (!p)
1180 return -EINVAL;
1181
1182#ifdef CONFIG_X86_32
1183 if (!strcmp(p, "nopentium")) {
1184 setup_clear_cpu_cap(X86_FEATURE_PSE);
1185 return 0;
1186 }
1187#endif
1188
1189 userdef = 1;
1190 mem_size = memparse(p, &p);
1191 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1192
1193 return 0;
1194}
1195early_param("mem", parse_memopt);
1196
1197static int __init parse_memmap_opt(char *p)
1198{
1199 char *oldp;
1200 u64 start_at, mem_size;
1201
1202 if (!p)
1203 return -EINVAL;
1204
1205 if (!strcmp(p, "exactmap")) {
1206#ifdef CONFIG_CRASH_DUMP
1207 /*
1208 * If we are doing a crash dump, we still need to know
1209 * the real mem size before original memory map is
1210 * reset.
1211 */
1212 saved_max_pfn = e820_end_of_ram_pfn();
1213#endif
1214 e820.nr_map = 0;
1215 userdef = 1;
1216 return 0;
1217 }
1218
1219 oldp = p;
1220 mem_size = memparse(p, &p);
1221 if (p == oldp)
1222 return -EINVAL;
1223
1224 userdef = 1;
1225 if (*p == '@') {
1226 start_at = memparse(p+1, &p);
1227 e820_add_region(start_at, mem_size, E820_RAM);
1228 } else if (*p == '#') {
1229 start_at = memparse(p+1, &p);
1230 e820_add_region(start_at, mem_size, E820_ACPI);
1231 } else if (*p == '$') {
1232 start_at = memparse(p+1, &p);
1233 e820_add_region(start_at, mem_size, E820_RESERVED);
1234 } else
1235 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1236
1237 return *p == '\0' ? 0 : -EINVAL;
1238}
1239early_param("memmap", parse_memmap_opt);
1240
1241void __init finish_e820_parsing(void)
1242{
1243 if (userdef) {
1244 int nr = e820.nr_map;
1245
1246 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
1247 early_panic("Invalid user supplied memory map");
1248 e820.nr_map = nr;
1249
1250 printk(KERN_INFO "user-defined physical RAM map:\n");
1251 e820_print_map("user");
1252 }
1253}
1254
1255static inline const char *e820_type_to_string(int e820_type)
1256{
1257 switch (e820_type) {
1258 case E820_RESERVED_KERN:
1259 case E820_RAM: return "System RAM";
1260 case E820_ACPI: return "ACPI Tables";
1261 case E820_NVS: return "ACPI Non-volatile Storage";
1262 default: return "reserved";
1263 }
1264}
1265
1266/*
1267 * Mark e820 reserved areas as busy for the resource manager.
1268 */
1269void __init e820_reserve_resources(void)
1270{
1271 int i;
1272 struct resource *res;
1273 u64 end;
1274
1275 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
1276 for (i = 0; i < e820.nr_map; i++) {
1277 end = e820.map[i].addr + e820.map[i].size - 1;
1278#ifndef CONFIG_RESOURCES_64BIT
1279 if (end > 0x100000000ULL) {
1280 res++;
1281 continue;
1282 }
1283#endif
1284 res->name = e820_type_to_string(e820.map[i].type);
1285 res->start = e820.map[i].addr;
1286 res->end = end;
1287
1288 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1289 insert_resource(&iomem_resource, res);
1290 res++;
1291 }
1292
1293 for (i = 0; i < e820_saved.nr_map; i++) {
1294 struct e820entry *entry = &e820_saved.map[i];
1295 firmware_map_add_early(entry->addr,
1296 entry->addr + entry->size - 1,
1297 e820_type_to_string(entry->type));
1298 }
1299}
1300
1301/*
1302 * Non-standard memory setup can be specified via this quirk:
1303 */
1304char * (*arch_memory_setup_quirk)(void);
1305
1306char *__init default_machine_specific_memory_setup(void)
1307{
1308 char *who = "BIOS-e820";
1309 int new_nr;
1310 /*
1311 * Try to copy the BIOS-supplied E820-map.
1312 *
1313 * Otherwise fake a memory map; one section from 0k->640k,
1314 * the next section from 1mb->appropriate_mem_k
1315 */
1316 new_nr = boot_params.e820_entries;
1317 sanitize_e820_map(boot_params.e820_map,
1318 ARRAY_SIZE(boot_params.e820_map),
1319 &new_nr);
1320 boot_params.e820_entries = new_nr;
1321 if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1322 < 0) {
1323 u64 mem_size;
1324
1325 /* compare results from other methods and take the greater */
1326 if (boot_params.alt_mem_k
1327 < boot_params.screen_info.ext_mem_k) {
1328 mem_size = boot_params.screen_info.ext_mem_k;
1329 who = "BIOS-88";
1330 } else {
1331 mem_size = boot_params.alt_mem_k;
1332 who = "BIOS-e801";
1333 }
1334
1335 e820.nr_map = 0;
1336 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1337 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1338 }
1339
1340 /* In case someone cares... */
1341 return who;
1342}
1343
1344char *__init __attribute__((weak)) machine_specific_memory_setup(void)
1345{
1346 if (arch_memory_setup_quirk) {
1347 char *who = arch_memory_setup_quirk();
1348
1349 if (who)
1350 return who;
1351 }
1352 return default_machine_specific_memory_setup();
1353}
1354
1355/* Overridden in paravirt.c if CONFIG_PARAVIRT */
1356char * __init __attribute__((weak)) memory_setup(void)
1357{
1358 return machine_specific_memory_setup();
1359}
1360
1361void __init setup_memory_map(void)
1362{
1363 char *who;
1364
1365 who = memory_setup();
1366 memcpy(&e820_saved, &e820, sizeof(struct e820map));
1367 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1368 e820_print_map(who);
1369}
1370
1371#ifdef CONFIG_X86_64
1372int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
1373{
1374 int i;
1375
1376 if (slot < 0 || slot >= e820.nr_map)
1377 return -1;
1378 for (i = slot; i < e820.nr_map; i++) {
1379 if (e820.map[i].type != E820_RAM)
1380 continue;
1381 break;
1382 }
1383 if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
1384 return -1;
1385 *addr = e820.map[i].addr;
1386 *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
1387 max_pfn << PAGE_SHIFT) - *addr;
1388 return i + 1;
1389}
1390#endif
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-05 11:31:17 -0500