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 CRC32
35 * checksum 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 its main 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(table: 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(table: 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, type: 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(table: 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(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
308static bool e820_nomerge(enum e820_type type)
309{
310 /*
311 * These types may indicate distinct platform ranges aligned to
312 * numa node, protection domain, performance domain, or other
313 * boundaries. Do not merge them.
314 */
315 if (type == E820_TYPE_PRAM)
316 return true;
317 if (type == E820_TYPE_SOFT_RESERVED)
318 return true;
319 return false;
320}
321
322int __init e820__update_table(struct e820_table *table)
323{
324 struct e820_entry *entries = table->entries;
325 u32 max_nr_entries = ARRAY_SIZE(table->entries);
326 enum e820_type current_type, last_type;
327 unsigned long long last_addr;
328 u32 new_nr_entries, overlap_entries;
329 u32 i, chg_idx, chg_nr;
330
331 /* If there's only one memory region, don't bother: */
332 if (table->nr_entries < 2)
333 return -1;
334
335 BUG_ON(table->nr_entries > max_nr_entries);
336
337 /* Bail out if we find any unreasonable addresses in the map: */
338 for (i = 0; i < table->nr_entries; i++) {
339 if (entries[i].addr + entries[i].size < entries[i].addr)
340 return -1;
341 }
342
343 /* Create pointers for initial change-point information (for sorting): */
344 for (i = 0; i < 2 * table->nr_entries; i++)
345 change_point[i] = &change_point_list[i];
346
347 /*
348 * Record all known change-points (starting and ending addresses),
349 * omitting empty memory regions:
350 */
351 chg_idx = 0;
352 for (i = 0; i < table->nr_entries; i++) {
353 if (entries[i].size != 0) {
354 change_point[chg_idx]->addr = entries[i].addr;
355 change_point[chg_idx++]->entry = &entries[i];
356 change_point[chg_idx]->addr = entries[i].addr + entries[i].size;
357 change_point[chg_idx++]->entry = &entries[i];
358 }
359 }
360 chg_nr = chg_idx;
361
362 /* Sort change-point list by memory addresses (low -> high): */
363 sort(base: change_point, num: chg_nr, size: sizeof(*change_point), cmp_func: cpcompare, NULL);
364
365 /* Create a new memory map, removing overlaps: */
366 overlap_entries = 0; /* Number of entries in the overlap table */
367 new_nr_entries = 0; /* Index for creating new map entries */
368 last_type = 0; /* Start with undefined memory type */
369 last_addr = 0; /* Start with 0 as last starting address */
370
371 /* Loop through change-points, determining effect on the new map: */
372 for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
373 /* Keep track of all overlapping entries */
374 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
375 /* Add map entry to overlap list (> 1 entry implies an overlap) */
376 overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
377 } else {
378 /* Remove entry from list (order independent, so swap with last): */
379 for (i = 0; i < overlap_entries; i++) {
380 if (overlap_list[i] == change_point[chg_idx]->entry)
381 overlap_list[i] = overlap_list[overlap_entries-1];
382 }
383 overlap_entries--;
384 }
385 /*
386 * If there are overlapping entries, decide which
387 * "type" to use (larger value takes precedence --
388 * 1=usable, 2,3,4,4+=unusable)
389 */
390 current_type = 0;
391 for (i = 0; i < overlap_entries; i++) {
392 if (overlap_list[i]->type > current_type)
393 current_type = overlap_list[i]->type;
394 }
395
396 /* Continue building up new map based on this information: */
397 if (current_type != last_type || e820_nomerge(type: current_type)) {
398 if (last_type) {
399 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
400 /* Move forward only if the new size was non-zero: */
401 if (new_entries[new_nr_entries].size != 0)
402 /* No more space left for new entries? */
403 if (++new_nr_entries >= max_nr_entries)
404 break;
405 }
406 if (current_type) {
407 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
408 new_entries[new_nr_entries].type = current_type;
409 last_addr = change_point[chg_idx]->addr;
410 }
411 last_type = current_type;
412 }
413 }
414
415 /* Copy the new entries into the original location: */
416 memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
417 table->nr_entries = new_nr_entries;
418
419 return 0;
420}
421
422static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
423{
424 struct boot_e820_entry *entry = entries;
425
426 while (nr_entries) {
427 u64 start = entry->addr;
428 u64 size = entry->size;
429 u64 end = start + size - 1;
430 u32 type = entry->type;
431
432 /* Ignore the entry on 64-bit overflow: */
433 if (start > end && likely(size))
434 return -1;
435
436 e820__range_add(start, size, type);
437
438 entry++;
439 nr_entries--;
440 }
441 return 0;
442}
443
444/*
445 * Copy the BIOS E820 map into a safe place.
446 *
447 * Sanity-check it while we're at it..
448 *
449 * If we're lucky and live on a modern system, the setup code
450 * will have given us a memory map that we can use to properly
451 * set up memory. If we aren't, we'll fake a memory map.
452 */
453static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
454{
455 /* Only one memory region (or negative)? Ignore it */
456 if (nr_entries < 2)
457 return -1;
458
459 return __append_e820_table(entries, nr_entries);
460}
461
462static u64 __init
463__e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
464{
465 u64 end;
466 unsigned int i;
467 u64 real_updated_size = 0;
468
469 BUG_ON(old_type == new_type);
470
471 if (size > (ULLONG_MAX - start))
472 size = ULLONG_MAX - start;
473
474 end = start + size;
475 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
476 e820_print_type(type: old_type);
477 pr_cont(" ==> ");
478 e820_print_type(type: new_type);
479 pr_cont("\n");
480
481 for (i = 0; i < table->nr_entries; i++) {
482 struct e820_entry *entry = &table->entries[i];
483 u64 final_start, final_end;
484 u64 entry_end;
485
486 if (entry->type != old_type)
487 continue;
488
489 entry_end = entry->addr + entry->size;
490
491 /* Completely covered by new range? */
492 if (entry->addr >= start && entry_end <= end) {
493 entry->type = new_type;
494 real_updated_size += entry->size;
495 continue;
496 }
497
498 /* New range is completely covered? */
499 if (entry->addr < start && entry_end > end) {
500 __e820__range_add(table, start, size, type: new_type);
501 __e820__range_add(table, start: end, size: entry_end - end, type: entry->type);
502 entry->size = start - entry->addr;
503 real_updated_size += size;
504 continue;
505 }
506
507 /* Partially covered: */
508 final_start = max(start, entry->addr);
509 final_end = min(end, entry_end);
510 if (final_start >= final_end)
511 continue;
512
513 __e820__range_add(table, start: final_start, size: final_end - final_start, type: new_type);
514
515 real_updated_size += final_end - final_start;
516
517 /*
518 * Left range could be head or tail, so need to update
519 * its size first:
520 */
521 entry->size -= final_end - final_start;
522 if (entry->addr < final_start)
523 continue;
524
525 entry->addr = final_end;
526 }
527 return real_updated_size;
528}
529
530u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
531{
532 return __e820__range_update(table: e820_table, start, size, old_type, new_type);
533}
534
535static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
536{
537 return __e820__range_update(table: e820_table_kexec, start, size, old_type, new_type);
538}
539
540/* Remove a range of memory from the E820 table: */
541u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
542{
543 int i;
544 u64 end;
545 u64 real_removed_size = 0;
546
547 if (size > (ULLONG_MAX - start))
548 size = ULLONG_MAX - start;
549
550 end = start + size;
551 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
552 if (check_type)
553 e820_print_type(type: old_type);
554 pr_cont("\n");
555
556 for (i = 0; i < e820_table->nr_entries; i++) {
557 struct e820_entry *entry = &e820_table->entries[i];
558 u64 final_start, final_end;
559 u64 entry_end;
560
561 if (check_type && entry->type != old_type)
562 continue;
563
564 entry_end = entry->addr + entry->size;
565
566 /* Completely covered? */
567 if (entry->addr >= start && entry_end <= end) {
568 real_removed_size += entry->size;
569 memset(entry, 0, sizeof(*entry));
570 continue;
571 }
572
573 /* Is the new range completely covered? */
574 if (entry->addr < start && entry_end > end) {
575 e820__range_add(start: end, size: entry_end - end, type: entry->type);
576 entry->size = start - entry->addr;
577 real_removed_size += size;
578 continue;
579 }
580
581 /* Partially covered: */
582 final_start = max(start, entry->addr);
583 final_end = min(end, entry_end);
584 if (final_start >= final_end)
585 continue;
586
587 real_removed_size += final_end - final_start;
588
589 /*
590 * Left range could be head or tail, so need to update
591 * the size first:
592 */
593 entry->size -= final_end - final_start;
594 if (entry->addr < final_start)
595 continue;
596
597 entry->addr = final_end;
598 }
599 return real_removed_size;
600}
601
602void __init e820__update_table_print(void)
603{
604 if (e820__update_table(table: e820_table))
605 return;
606
607 pr_info("modified physical RAM map:\n");
608 e820__print_table(who: "modified");
609}
610
611static void __init e820__update_table_kexec(void)
612{
613 e820__update_table(table: e820_table_kexec);
614}
615
616#define MAX_GAP_END 0x100000000ull
617
618/*
619 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
620 */
621static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
622{
623 unsigned long long last = MAX_GAP_END;
624 int i = e820_table->nr_entries;
625 int found = 0;
626
627 while (--i >= 0) {
628 unsigned long long start = e820_table->entries[i].addr;
629 unsigned long long end = start + e820_table->entries[i].size;
630
631 /*
632 * Since "last" is at most 4GB, we know we'll
633 * fit in 32 bits if this condition is true:
634 */
635 if (last > end) {
636 unsigned long gap = last - end;
637
638 if (gap >= *gapsize) {
639 *gapsize = gap;
640 *gapstart = end;
641 found = 1;
642 }
643 }
644 if (start < last)
645 last = start;
646 }
647 return found;
648}
649
650/*
651 * Search for the biggest gap in the low 32 bits of the E820
652 * memory space. We pass this space to the PCI subsystem, so
653 * that it can assign MMIO resources for hotplug or
654 * unconfigured devices in.
655 *
656 * Hopefully the BIOS let enough space left.
657 */
658__init void e820__setup_pci_gap(void)
659{
660 unsigned long gapstart, gapsize;
661 int found;
662
663 gapsize = 0x400000;
664 found = e820_search_gap(gapstart: &gapstart, gapsize: &gapsize);
665
666 if (!found) {
667#ifdef CONFIG_X86_64
668 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
669 pr_err("Cannot find an available gap in the 32-bit address range\n");
670 pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
671#else
672 gapstart = 0x10000000;
673#endif
674 }
675
676 /*
677 * e820__reserve_resources_late() protects stolen RAM already:
678 */
679 pci_mem_start = gapstart;
680
681 pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
682 gapstart, gapstart + gapsize - 1);
683}
684
685/*
686 * Called late during init, in free_initmem().
687 *
688 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
689 *
690 * Copy them to a (usually much smaller) dynamically allocated area that is
691 * sized precisely after the number of e820 entries.
692 *
693 * This is done after we've performed all the fixes and tweaks to the tables.
694 * All functions which modify them are __init functions, which won't exist
695 * after free_initmem().
696 */
697__init void e820__reallocate_tables(void)
698{
699 struct e820_table *n;
700 int size;
701
702 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
703 n = kmemdup(p: e820_table, size, GFP_KERNEL);
704 BUG_ON(!n);
705 e820_table = n;
706
707 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
708 n = kmemdup(p: e820_table_kexec, size, GFP_KERNEL);
709 BUG_ON(!n);
710 e820_table_kexec = n;
711
712 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
713 n = kmemdup(p: e820_table_firmware, size, GFP_KERNEL);
714 BUG_ON(!n);
715 e820_table_firmware = n;
716}
717
718/*
719 * Because of the small fixed size of struct boot_params, only the first
720 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
721 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
722 * struct setup_data, which is parsed here.
723 */
724void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
725{
726 int entries;
727 struct boot_e820_entry *extmap;
728 struct setup_data *sdata;
729
730 sdata = early_memremap(phys_addr, size: data_len);
731 entries = sdata->len / sizeof(*extmap);
732 extmap = (struct boot_e820_entry *)(sdata->data);
733
734 __append_e820_table(entries: extmap, nr_entries: entries);
735 e820__update_table(table: e820_table);
736
737 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
738 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
739
740 early_memunmap(addr: sdata, size: data_len);
741 pr_info("extended physical RAM map:\n");
742 e820__print_table(who: "extended");
743}
744
745/*
746 * Find the ranges of physical addresses that do not correspond to
747 * E820 RAM areas and register the corresponding pages as 'nosave' for
748 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
749 *
750 * This function requires the E820 map to be sorted and without any
751 * overlapping entries.
752 */
753void __init e820__register_nosave_regions(unsigned long limit_pfn)
754{
755 int i;
756 unsigned long pfn = 0;
757
758 for (i = 0; i < e820_table->nr_entries; i++) {
759 struct e820_entry *entry = &e820_table->entries[i];
760
761 if (pfn < PFN_UP(entry->addr))
762 register_nosave_region(b: pfn, PFN_UP(entry->addr));
763
764 pfn = PFN_DOWN(entry->addr + entry->size);
765
766 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
767 register_nosave_region(PFN_UP(entry->addr), e: pfn);
768
769 if (pfn >= limit_pfn)
770 break;
771 }
772}
773
774#ifdef CONFIG_ACPI
775/*
776 * Register ACPI NVS memory regions, so that we can save/restore them during
777 * hibernation and the subsequent resume:
778 */
779static int __init e820__register_nvs_regions(void)
780{
781 int i;
782
783 for (i = 0; i < e820_table->nr_entries; i++) {
784 struct e820_entry *entry = &e820_table->entries[i];
785
786 if (entry->type == E820_TYPE_NVS)
787 acpi_nvs_register(start: entry->addr, size: entry->size);
788 }
789
790 return 0;
791}
792core_initcall(e820__register_nvs_regions);
793#endif
794
795/*
796 * Allocate the requested number of bytes with the requested alignment
797 * and return (the physical address) to the caller. Also register this
798 * range in the 'kexec' E820 table as a reserved range.
799 *
800 * This allows kexec to fake a new mptable, as if it came from the real
801 * system.
802 */
803u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
804{
805 u64 addr;
806
807 addr = memblock_phys_alloc(size, align);
808 if (addr) {
809 e820__range_update_kexec(start: addr, size, old_type: E820_TYPE_RAM, new_type: E820_TYPE_RESERVED);
810 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
811 e820__update_table_kexec();
812 }
813
814 return addr;
815}
816
817#ifdef CONFIG_X86_32
818# ifdef CONFIG_X86_PAE
819# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
820# else
821# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
822# endif
823#else /* CONFIG_X86_32 */
824# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
825#endif
826
827/*
828 * Find the highest page frame number we have available
829 */
830static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
831{
832 int i;
833 unsigned long last_pfn = 0;
834 unsigned long max_arch_pfn = MAX_ARCH_PFN;
835
836 for (i = 0; i < e820_table->nr_entries; i++) {
837 struct e820_entry *entry = &e820_table->entries[i];
838 unsigned long start_pfn;
839 unsigned long end_pfn;
840
841 if (entry->type != type)
842 continue;
843
844 start_pfn = entry->addr >> PAGE_SHIFT;
845 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
846
847 if (start_pfn >= limit_pfn)
848 continue;
849 if (end_pfn > limit_pfn) {
850 last_pfn = limit_pfn;
851 break;
852 }
853 if (end_pfn > last_pfn)
854 last_pfn = end_pfn;
855 }
856
857 if (last_pfn > max_arch_pfn)
858 last_pfn = max_arch_pfn;
859
860 pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
861 last_pfn, max_arch_pfn);
862 return last_pfn;
863}
864
865unsigned long __init e820__end_of_ram_pfn(void)
866{
867 return e820_end_pfn(MAX_ARCH_PFN, type: E820_TYPE_RAM);
868}
869
870unsigned long __init e820__end_of_low_ram_pfn(void)
871{
872 return e820_end_pfn(limit_pfn: 1UL << (32 - PAGE_SHIFT), type: E820_TYPE_RAM);
873}
874
875static void __init early_panic(char *msg)
876{
877 early_printk(fmt: msg);
878 panic(fmt: msg);
879}
880
881static int userdef __initdata;
882
883/* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
884static int __init parse_memopt(char *p)
885{
886 u64 mem_size;
887
888 if (!p)
889 return -EINVAL;
890
891 if (!strcmp(p, "nopentium")) {
892#ifdef CONFIG_X86_32
893 setup_clear_cpu_cap(X86_FEATURE_PSE);
894 return 0;
895#else
896 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
897 return -EINVAL;
898#endif
899 }
900
901 userdef = 1;
902 mem_size = memparse(ptr: p, retptr: &p);
903
904 /* Don't remove all memory when getting "mem={invalid}" parameter: */
905 if (mem_size == 0)
906 return -EINVAL;
907
908 e820__range_remove(start: mem_size, ULLONG_MAX - mem_size, old_type: E820_TYPE_RAM, check_type: 1);
909
910#ifdef CONFIG_MEMORY_HOTPLUG
911 max_mem_size = mem_size;
912#endif
913
914 return 0;
915}
916early_param("mem", parse_memopt);
917
918static int __init parse_memmap_one(char *p)
919{
920 char *oldp;
921 u64 start_at, mem_size;
922
923 if (!p)
924 return -EINVAL;
925
926 if (!strncmp(p, "exactmap", 8)) {
927 e820_table->nr_entries = 0;
928 userdef = 1;
929 return 0;
930 }
931
932 oldp = p;
933 mem_size = memparse(ptr: p, retptr: &p);
934 if (p == oldp)
935 return -EINVAL;
936
937 userdef = 1;
938 if (*p == '@') {
939 start_at = memparse(ptr: p+1, retptr: &p);
940 e820__range_add(start: start_at, size: mem_size, type: E820_TYPE_RAM);
941 } else if (*p == '#') {
942 start_at = memparse(ptr: p+1, retptr: &p);
943 e820__range_add(start: start_at, size: mem_size, type: E820_TYPE_ACPI);
944 } else if (*p == '$') {
945 start_at = memparse(ptr: p+1, retptr: &p);
946 e820__range_add(start: start_at, size: mem_size, type: E820_TYPE_RESERVED);
947 } else if (*p == '!') {
948 start_at = memparse(ptr: p+1, retptr: &p);
949 e820__range_add(start: start_at, size: mem_size, type: E820_TYPE_PRAM);
950 } else if (*p == '%') {
951 enum e820_type from = 0, to = 0;
952
953 start_at = memparse(ptr: p + 1, retptr: &p);
954 if (*p == '-')
955 from = simple_strtoull(p + 1, &p, 0);
956 if (*p == '+')
957 to = simple_strtoull(p + 1, &p, 0);
958 if (*p != '\0')
959 return -EINVAL;
960 if (from && to)
961 e820__range_update(start: start_at, size: mem_size, old_type: from, new_type: to);
962 else if (to)
963 e820__range_add(start: start_at, size: mem_size, type: to);
964 else if (from)
965 e820__range_remove(start: start_at, size: mem_size, old_type: from, check_type: 1);
966 else
967 e820__range_remove(start: start_at, size: mem_size, old_type: 0, check_type: 0);
968 } else {
969 e820__range_remove(start: mem_size, ULLONG_MAX - mem_size, old_type: E820_TYPE_RAM, check_type: 1);
970 }
971
972 return *p == '\0' ? 0 : -EINVAL;
973}
974
975static int __init parse_memmap_opt(char *str)
976{
977 while (str) {
978 char *k = strchr(str, ',');
979
980 if (k)
981 *k++ = 0;
982
983 parse_memmap_one(p: str);
984 str = k;
985 }
986
987 return 0;
988}
989early_param("memmap", parse_memmap_opt);
990
991/*
992 * Reserve all entries from the bootloader's extensible data nodes list,
993 * because if present we are going to use it later on to fetch e820
994 * entries from it:
995 */
996void __init e820__reserve_setup_data(void)
997{
998 struct setup_indirect *indirect;
999 struct setup_data *data;
1000 u64 pa_data, pa_next;
1001 u32 len;
1002
1003 pa_data = boot_params.hdr.setup_data;
1004 if (!pa_data)
1005 return;
1006
1007 while (pa_data) {
1008 data = early_memremap(phys_addr: pa_data, size: sizeof(*data));
1009 if (!data) {
1010 pr_warn("e820: failed to memremap setup_data entry\n");
1011 return;
1012 }
1013
1014 len = sizeof(*data);
1015 pa_next = data->next;
1016
1017 e820__range_update(start: pa_data, size: sizeof(*data)+data->len, old_type: E820_TYPE_RAM, new_type: E820_TYPE_RESERVED_KERN);
1018
1019 /*
1020 * SETUP_EFI and SETUP_IMA are supplied by kexec and do not need
1021 * to be reserved.
1022 */
1023 if (data->type != SETUP_EFI && data->type != SETUP_IMA)
1024 e820__range_update_kexec(start: pa_data,
1025 size: sizeof(*data) + data->len,
1026 old_type: E820_TYPE_RAM, new_type: E820_TYPE_RESERVED_KERN);
1027
1028 if (data->type == SETUP_INDIRECT) {
1029 len += data->len;
1030 early_memunmap(addr: data, size: sizeof(*data));
1031 data = early_memremap(phys_addr: pa_data, size: len);
1032 if (!data) {
1033 pr_warn("e820: failed to memremap indirect setup_data\n");
1034 return;
1035 }
1036
1037 indirect = (struct setup_indirect *)data->data;
1038
1039 if (indirect->type != SETUP_INDIRECT) {
1040 e820__range_update(start: indirect->addr, size: indirect->len,
1041 old_type: E820_TYPE_RAM, new_type: E820_TYPE_RESERVED_KERN);
1042 e820__range_update_kexec(start: indirect->addr, size: indirect->len,
1043 old_type: E820_TYPE_RAM, new_type: E820_TYPE_RESERVED_KERN);
1044 }
1045 }
1046
1047 pa_data = pa_next;
1048 early_memunmap(addr: data, size: len);
1049 }
1050
1051 e820__update_table(table: e820_table);
1052 e820__update_table(table: e820_table_kexec);
1053
1054 pr_info("extended physical RAM map:\n");
1055 e820__print_table(who: "reserve setup_data");
1056}
1057
1058/*
1059 * Called after parse_early_param(), after early parameters (such as mem=)
1060 * have been processed, in which case we already have an E820 table filled in
1061 * via the parameter callback function(s), but it's not sorted and printed yet:
1062 */
1063void __init e820__finish_early_params(void)
1064{
1065 if (userdef) {
1066 if (e820__update_table(table: e820_table) < 0)
1067 early_panic(msg: "Invalid user supplied memory map");
1068
1069 pr_info("user-defined physical RAM map:\n");
1070 e820__print_table(who: "user");
1071 }
1072}
1073
1074static const char *__init e820_type_to_string(struct e820_entry *entry)
1075{
1076 switch (entry->type) {
1077 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1078 case E820_TYPE_RAM: return "System RAM";
1079 case E820_TYPE_ACPI: return "ACPI Tables";
1080 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1081 case E820_TYPE_UNUSABLE: return "Unusable memory";
1082 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1083 case E820_TYPE_PMEM: return "Persistent Memory";
1084 case E820_TYPE_RESERVED: return "Reserved";
1085 case E820_TYPE_SOFT_RESERVED: return "Soft Reserved";
1086 default: return "Unknown E820 type";
1087 }
1088}
1089
1090static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1091{
1092 switch (entry->type) {
1093 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1094 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1095 case E820_TYPE_ACPI: /* Fall-through: */
1096 case E820_TYPE_NVS: /* Fall-through: */
1097 case E820_TYPE_UNUSABLE: /* Fall-through: */
1098 case E820_TYPE_PRAM: /* Fall-through: */
1099 case E820_TYPE_PMEM: /* Fall-through: */
1100 case E820_TYPE_RESERVED: /* Fall-through: */
1101 case E820_TYPE_SOFT_RESERVED: /* Fall-through: */
1102 default: return IORESOURCE_MEM;
1103 }
1104}
1105
1106static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1107{
1108 switch (entry->type) {
1109 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1110 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1111 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1112 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1113 case E820_TYPE_RESERVED: return IORES_DESC_RESERVED;
1114 case E820_TYPE_SOFT_RESERVED: return IORES_DESC_SOFT_RESERVED;
1115 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1116 case E820_TYPE_RAM: /* Fall-through: */
1117 case E820_TYPE_UNUSABLE: /* Fall-through: */
1118 default: return IORES_DESC_NONE;
1119 }
1120}
1121
1122static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1123{
1124 /* this is the legacy bios/dos rom-shadow + mmio region */
1125 if (res->start < (1ULL<<20))
1126 return true;
1127
1128 /*
1129 * Treat persistent memory and other special memory ranges like
1130 * device memory, i.e. reserve it for exclusive use of a driver
1131 */
1132 switch (type) {
1133 case E820_TYPE_RESERVED:
1134 case E820_TYPE_SOFT_RESERVED:
1135 case E820_TYPE_PRAM:
1136 case E820_TYPE_PMEM:
1137 return false;
1138 case E820_TYPE_RESERVED_KERN:
1139 case E820_TYPE_RAM:
1140 case E820_TYPE_ACPI:
1141 case E820_TYPE_NVS:
1142 case E820_TYPE_UNUSABLE:
1143 default:
1144 return true;
1145 }
1146}
1147
1148/*
1149 * Mark E820 reserved areas as busy for the resource manager:
1150 */
1151
1152static struct resource __initdata *e820_res;
1153
1154void __init e820__reserve_resources(void)
1155{
1156 int i;
1157 struct resource *res;
1158 u64 end;
1159
1160 res = memblock_alloc(size: sizeof(*res) * e820_table->nr_entries,
1161 SMP_CACHE_BYTES);
1162 if (!res)
1163 panic(fmt: "%s: Failed to allocate %zu bytes\n", __func__,
1164 sizeof(*res) * e820_table->nr_entries);
1165 e820_res = res;
1166
1167 for (i = 0; i < e820_table->nr_entries; i++) {
1168 struct e820_entry *entry = e820_table->entries + i;
1169
1170 end = entry->addr + entry->size - 1;
1171 if (end != (resource_size_t)end) {
1172 res++;
1173 continue;
1174 }
1175 res->start = entry->addr;
1176 res->end = end;
1177 res->name = e820_type_to_string(entry);
1178 res->flags = e820_type_to_iomem_type(entry);
1179 res->desc = e820_type_to_iores_desc(entry);
1180
1181 /*
1182 * Don't register the region that could be conflicted with
1183 * PCI device BAR resources and insert them later in
1184 * pcibios_resource_survey():
1185 */
1186 if (do_mark_busy(type: entry->type, res)) {
1187 res->flags |= IORESOURCE_BUSY;
1188 insert_resource(parent: &iomem_resource, new: res);
1189 }
1190 res++;
1191 }
1192
1193 /* Expose the bootloader-provided memory layout to the sysfs. */
1194 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1195 struct e820_entry *entry = e820_table_firmware->entries + i;
1196
1197 firmware_map_add_early(start: entry->addr, end: entry->addr + entry->size, type: e820_type_to_string(entry));
1198 }
1199}
1200
1201/*
1202 * How much should we pad the end of RAM, depending on where it is?
1203 */
1204static unsigned long __init ram_alignment(resource_size_t pos)
1205{
1206 unsigned long mb = pos >> 20;
1207
1208 /* To 64kB in the first megabyte */
1209 if (!mb)
1210 return 64*1024;
1211
1212 /* To 1MB in the first 16MB */
1213 if (mb < 16)
1214 return 1024*1024;
1215
1216 /* To 64MB for anything above that */
1217 return 64*1024*1024;
1218}
1219
1220#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1221
1222void __init e820__reserve_resources_late(void)
1223{
1224 int i;
1225 struct resource *res;
1226
1227 res = e820_res;
1228 for (i = 0; i < e820_table->nr_entries; i++) {
1229 if (!res->parent && res->end)
1230 insert_resource_expand_to_fit(root: &iomem_resource, new: res);
1231 res++;
1232 }
1233
1234 /*
1235 * Try to bump up RAM regions to reasonable boundaries, to
1236 * avoid stolen RAM:
1237 */
1238 for (i = 0; i < e820_table->nr_entries; i++) {
1239 struct e820_entry *entry = &e820_table->entries[i];
1240 u64 start, end;
1241
1242 if (entry->type != E820_TYPE_RAM)
1243 continue;
1244
1245 start = entry->addr + entry->size;
1246 end = round_up(start, ram_alignment(start)) - 1;
1247 if (end > MAX_RESOURCE_SIZE)
1248 end = MAX_RESOURCE_SIZE;
1249 if (start >= end)
1250 continue;
1251
1252 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1253 reserve_region_with_split(root: &iomem_resource, start, end, name: "RAM buffer");
1254 }
1255}
1256
1257/*
1258 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1259 */
1260char *__init e820__memory_setup_default(void)
1261{
1262 char *who = "BIOS-e820";
1263
1264 /*
1265 * Try to copy the BIOS-supplied E820-map.
1266 *
1267 * Otherwise fake a memory map; one section from 0k->640k,
1268 * the next section from 1mb->appropriate_mem_k
1269 */
1270 if (append_e820_table(entries: boot_params.e820_table, nr_entries: boot_params.e820_entries) < 0) {
1271 u64 mem_size;
1272
1273 /* Compare results from other methods and take the one that gives more RAM: */
1274 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1275 mem_size = boot_params.screen_info.ext_mem_k;
1276 who = "BIOS-88";
1277 } else {
1278 mem_size = boot_params.alt_mem_k;
1279 who = "BIOS-e801";
1280 }
1281
1282 e820_table->nr_entries = 0;
1283 e820__range_add(start: 0, LOWMEMSIZE(), type: E820_TYPE_RAM);
1284 e820__range_add(HIGH_MEMORY, size: mem_size << 10, type: E820_TYPE_RAM);
1285 }
1286
1287 /* We just appended a lot of ranges, sanitize the table: */
1288 e820__update_table(table: e820_table);
1289
1290 return who;
1291}
1292
1293/*
1294 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1295 * E820 map - with an optional platform quirk available for virtual platforms
1296 * to override this method of boot environment processing:
1297 */
1298void __init e820__memory_setup(void)
1299{
1300 char *who;
1301
1302 /* This is a firmware interface ABI - make sure we don't break it: */
1303 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1304
1305 who = x86_init.resources.memory_setup();
1306
1307 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1308 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1309
1310 pr_info("BIOS-provided physical RAM map:\n");
1311 e820__print_table(who);
1312}
1313
1314void __init e820__memblock_setup(void)
1315{
1316 int i;
1317 u64 end;
1318
1319 /*
1320 * The bootstrap memblock region count maximum is 128 entries
1321 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1322 * than that - so allow memblock resizing.
1323 *
1324 * This is safe, because this call happens pretty late during x86 setup,
1325 * so we know about reserved memory regions already. (This is important
1326 * so that memblock resizing does no stomp over reserved areas.)
1327 */
1328 memblock_allow_resize();
1329
1330 for (i = 0; i < e820_table->nr_entries; i++) {
1331 struct e820_entry *entry = &e820_table->entries[i];
1332
1333 end = entry->addr + entry->size;
1334 if (end != (resource_size_t)end)
1335 continue;
1336
1337 if (entry->type == E820_TYPE_SOFT_RESERVED)
1338 memblock_reserve(base: entry->addr, size: entry->size);
1339
1340 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1341 continue;
1342
1343 memblock_add(base: entry->addr, size: entry->size);
1344 }
1345
1346 /* Throw away partial pages: */
1347 memblock_trim_memory(PAGE_SIZE);
1348
1349 memblock_dump_all();
1350}
1351

source code of linux/arch/x86/kernel/e820.c