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 * - the hibernation code uses it to generate a kernel-independent CRC32
32 * checksum of the physical memory layout of a system.
33 *
34 * - 'e820_table_kexec': a slightly modified (by the kernel) firmware version
35 * passed to us by the bootloader - the major difference between
36 * e820_table_firmware[] and this one is that e820_table_kexec[]
37 * might be modified by the kexec itself to fake an mptable.
38 * We use this to:
39 *
40 * - kexec, which is a bootloader in disguise, uses the original E820
41 * layout to pass to the kexec-ed kernel. This way the original kernel
42 * can have a restricted E820 map while the kexec()-ed kexec-kernel
43 * can have access to full memory - etc.
44 *
45 * Export the memory layout via /sys/firmware/memmap. kexec-tools uses
46 * the entries to create an E820 table for the kexec kernel.
47 *
48 * kexec_file_load in-kernel code uses the table for the kexec kernel.
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: pr_cont("usable"); break;
191 case E820_TYPE_RESERVED: pr_cont("reserved"); break;
192 case E820_TYPE_SOFT_RESERVED: pr_cont("soft reserved"); break;
193 case E820_TYPE_ACPI: pr_cont("ACPI data"); break;
194 case E820_TYPE_NVS: pr_cont("ACPI NVS"); break;
195 case E820_TYPE_UNUSABLE: pr_cont("unusable"); break;
196 case E820_TYPE_PMEM: /* Fall through: */
197 case E820_TYPE_PRAM: pr_cont("persistent (type %u)", type); break;
198 default: pr_cont("type %u", type); break;
199 }
200}
201
202void __init e820__print_table(char *who)
203{
204 int i;
205
206 for (i = 0; i < e820_table->nr_entries; i++) {
207 pr_info("%s: [mem %#018Lx-%#018Lx] ",
208 who,
209 e820_table->entries[i].addr,
210 e820_table->entries[i].addr + e820_table->entries[i].size - 1);
211
212 e820_print_type(type: e820_table->entries[i].type);
213 pr_cont("\n");
214 }
215}
216
217/*
218 * Sanitize an E820 map.
219 *
220 * Some E820 layouts include overlapping entries. The following
221 * replaces the original E820 map with a new one, removing overlaps,
222 * and resolving conflicting memory types in favor of highest
223 * numbered type.
224 *
225 * The input parameter 'entries' points to an array of 'struct
226 * e820_entry' which on entry has elements in the range [0, *nr_entries)
227 * valid, and which has space for up to max_nr_entries entries.
228 * On return, the resulting sanitized E820 map entries will be in
229 * overwritten in the same location, starting at 'entries'.
230 *
231 * The integer pointed to by nr_entries must be valid on entry (the
232 * current number of valid entries located at 'entries'). If the
233 * sanitizing succeeds the *nr_entries will be updated with the new
234 * number of valid entries (something no more than max_nr_entries).
235 *
236 * The return value from e820__update_table() is zero if it
237 * successfully 'sanitized' the map entries passed in, and is -1
238 * if it did nothing, which can happen if either of (1) it was
239 * only passed one map entry, or (2) any of the input map entries
240 * were invalid (start + size < start, meaning that the size was
241 * so big the described memory range wrapped around through zero.)
242 *
243 * Visually we're performing the following
244 * (1,2,3,4 = memory types)...
245 *
246 * Sample memory map (w/overlaps):
247 * ____22__________________
248 * ______________________4_
249 * ____1111________________
250 * _44_____________________
251 * 11111111________________
252 * ____________________33__
253 * ___________44___________
254 * __________33333_________
255 * ______________22________
256 * ___________________2222_
257 * _________111111111______
258 * _____________________11_
259 * _________________4______
260 *
261 * Sanitized equivalent (no overlap):
262 * 1_______________________
263 * _44_____________________
264 * ___1____________________
265 * ____22__________________
266 * ______11________________
267 * _________1______________
268 * __________3_____________
269 * ___________44___________
270 * _____________33_________
271 * _______________2________
272 * ________________1_______
273 * _________________4______
274 * ___________________2____
275 * ____________________33__
276 * ______________________4_
277 */
278struct change_member {
279 /* Pointer to the original entry: */
280 struct e820_entry *entry;
281 /* Address for this change point: */
282 unsigned long long addr;
283};
284
285static struct change_member change_point_list[2*E820_MAX_ENTRIES] __initdata;
286static struct change_member *change_point[2*E820_MAX_ENTRIES] __initdata;
287static struct e820_entry *overlap_list[E820_MAX_ENTRIES] __initdata;
288static struct e820_entry new_entries[E820_MAX_ENTRIES] __initdata;
289
290static int __init cpcompare(const void *a, const void *b)
291{
292 struct change_member * const *app = a, * const *bpp = b;
293 const struct change_member *ap = *app, *bp = *bpp;
294
295 /*
296 * Inputs are pointers to two elements of change_point[]. If their
297 * addresses are not equal, their difference dominates. If the addresses
298 * are equal, then consider one that represents the end of its region
299 * to be greater than one that does not.
300 */
301 if (ap->addr != bp->addr)
302 return ap->addr > bp->addr ? 1 : -1;
303
304 return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
305}
306
307static bool e820_nomerge(enum e820_type type)
308{
309 /*
310 * These types may indicate distinct platform ranges aligned to
311 * numa node, protection domain, performance domain, or other
312 * boundaries. Do not merge them.
313 */
314 if (type == E820_TYPE_PRAM)
315 return true;
316 if (type == E820_TYPE_SOFT_RESERVED)
317 return true;
318 return false;
319}
320
321int __init e820__update_table(struct e820_table *table)
322{
323 struct e820_entry *entries = table->entries;
324 u32 max_nr_entries = ARRAY_SIZE(table->entries);
325 enum e820_type current_type, last_type;
326 unsigned long long last_addr;
327 u32 new_nr_entries, overlap_entries;
328 u32 i, chg_idx, chg_nr;
329
330 /* If there's only one memory region, don't bother: */
331 if (table->nr_entries < 2)
332 return -1;
333
334 BUG_ON(table->nr_entries > max_nr_entries);
335
336 /* Bail out if we find any unreasonable addresses in the map: */
337 for (i = 0; i < table->nr_entries; i++) {
338 if (entries[i].addr + entries[i].size < entries[i].addr)
339 return -1;
340 }
341
342 /* Create pointers for initial change-point information (for sorting): */
343 for (i = 0; i < 2 * table->nr_entries; i++)
344 change_point[i] = &change_point_list[i];
345
346 /*
347 * Record all known change-points (starting and ending addresses),
348 * omitting empty memory regions:
349 */
350 chg_idx = 0;
351 for (i = 0; i < table->nr_entries; i++) {
352 if (entries[i].size != 0) {
353 change_point[chg_idx]->addr = entries[i].addr;
354 change_point[chg_idx++]->entry = &entries[i];
355 change_point[chg_idx]->addr = entries[i].addr + entries[i].size;
356 change_point[chg_idx++]->entry = &entries[i];
357 }
358 }
359 chg_nr = chg_idx;
360
361 /* Sort change-point list by memory addresses (low -> high): */
362 sort(base: change_point, num: chg_nr, size: sizeof(*change_point), cmp_func: cpcompare, NULL);
363
364 /* Create a new memory map, removing overlaps: */
365 overlap_entries = 0; /* Number of entries in the overlap table */
366 new_nr_entries = 0; /* Index for creating new map entries */
367 last_type = 0; /* Start with undefined memory type */
368 last_addr = 0; /* Start with 0 as last starting address */
369
370 /* Loop through change-points, determining effect on the new map: */
371 for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
372 /* Keep track of all overlapping entries */
373 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
374 /* Add map entry to overlap list (> 1 entry implies an overlap) */
375 overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
376 } else {
377 /* Remove entry from list (order independent, so swap with last): */
378 for (i = 0; i < overlap_entries; i++) {
379 if (overlap_list[i] == change_point[chg_idx]->entry)
380 overlap_list[i] = overlap_list[overlap_entries-1];
381 }
382 overlap_entries--;
383 }
384 /*
385 * If there are overlapping entries, decide which
386 * "type" to use (larger value takes precedence --
387 * 1=usable, 2,3,4,4+=unusable)
388 */
389 current_type = 0;
390 for (i = 0; i < overlap_entries; i++) {
391 if (overlap_list[i]->type > current_type)
392 current_type = overlap_list[i]->type;
393 }
394
395 /* Continue building up new map based on this information: */
396 if (current_type != last_type || e820_nomerge(type: current_type)) {
397 if (last_type) {
398 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
399 /* Move forward only if the new size was non-zero: */
400 if (new_entries[new_nr_entries].size != 0)
401 /* No more space left for new entries? */
402 if (++new_nr_entries >= max_nr_entries)
403 break;
404 }
405 if (current_type) {
406 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
407 new_entries[new_nr_entries].type = current_type;
408 last_addr = change_point[chg_idx]->addr;
409 }
410 last_type = current_type;
411 }
412 }
413
414 /* Copy the new entries into the original location: */
415 memcpy(to: entries, from: new_entries, len: new_nr_entries*sizeof(*entries));
416 table->nr_entries = new_nr_entries;
417
418 return 0;
419}
420
421static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
422{
423 struct boot_e820_entry *entry = entries;
424
425 while (nr_entries) {
426 u64 start = entry->addr;
427 u64 size = entry->size;
428 u64 end = start + size - 1;
429 u32 type = entry->type;
430
431 /* Ignore the entry on 64-bit overflow: */
432 if (start > end && likely(size))
433 return -1;
434
435 e820__range_add(start, size, type);
436
437 entry++;
438 nr_entries--;
439 }
440 return 0;
441}
442
443/*
444 * Copy the BIOS E820 map into a safe place.
445 *
446 * Sanity-check it while we're at it..
447 *
448 * If we're lucky and live on a modern system, the setup code
449 * will have given us a memory map that we can use to properly
450 * set up memory. If we aren't, we'll fake a memory map.
451 */
452static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
453{
454 /* Only one memory region (or negative)? Ignore it */
455 if (nr_entries < 2)
456 return -1;
457
458 return __append_e820_table(entries, nr_entries);
459}
460
461static u64 __init
462__e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
463{
464 u64 end;
465 unsigned int i;
466 u64 real_updated_size = 0;
467
468 BUG_ON(old_type == new_type);
469
470 if (size > (ULLONG_MAX - start))
471 size = ULLONG_MAX - start;
472
473 end = start + size;
474 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
475 e820_print_type(type: old_type);
476 pr_cont(" ==> ");
477 e820_print_type(type: new_type);
478 pr_cont("\n");
479
480 for (i = 0; i < table->nr_entries; i++) {
481 struct e820_entry *entry = &table->entries[i];
482 u64 final_start, final_end;
483 u64 entry_end;
484
485 if (entry->type != old_type)
486 continue;
487
488 entry_end = entry->addr + entry->size;
489
490 /* Completely covered by new range? */
491 if (entry->addr >= start && entry_end <= end) {
492 entry->type = new_type;
493 real_updated_size += entry->size;
494 continue;
495 }
496
497 /* New range is completely covered? */
498 if (entry->addr < start && entry_end > end) {
499 __e820__range_add(table, start, size, type: new_type);
500 __e820__range_add(table, start: end, size: entry_end - end, type: entry->type);
501 entry->size = start - entry->addr;
502 real_updated_size += size;
503 continue;
504 }
505
506 /* Partially covered: */
507 final_start = max(start, entry->addr);
508 final_end = min(end, entry_end);
509 if (final_start >= final_end)
510 continue;
511
512 __e820__range_add(table, start: final_start, size: final_end - final_start, type: new_type);
513
514 real_updated_size += final_end - final_start;
515
516 /*
517 * Left range could be head or tail, so need to update
518 * its size first:
519 */
520 entry->size -= final_end - final_start;
521 if (entry->addr < final_start)
522 continue;
523
524 entry->addr = final_end;
525 }
526 return real_updated_size;
527}
528
529u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
530{
531 return __e820__range_update(table: e820_table, start, size, old_type, new_type);
532}
533
534u64 __init e820__range_update_table(struct e820_table *t, u64 start, u64 size,
535 enum e820_type old_type, enum e820_type new_type)
536{
537 return __e820__range_update(table: t, 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(s: entry, c: 0, n: 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(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(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(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(to: e820_table_kexec, from: e820_table, len: sizeof(*e820_table_kexec));
738 memcpy(to: e820_table_firmware, from: e820_table, len: 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 u64 last_addr = 0;
757
758 for (i = 0; i < e820_table->nr_entries; i++) {
759 struct e820_entry *entry = &e820_table->entries[i];
760
761 if (entry->type != E820_TYPE_RAM)
762 continue;
763
764 if (last_addr < entry->addr)
765 register_nosave_region(PFN_DOWN(last_addr), PFN_UP(entry->addr));
766
767 last_addr = entry->addr + entry->size;
768 }
769
770 register_nosave_region(PFN_DOWN(last_addr), e: limit_pfn);
771}
772
773#ifdef CONFIG_ACPI
774/*
775 * Register ACPI NVS memory regions, so that we can save/restore them during
776 * hibernation and the subsequent resume:
777 */
778static int __init e820__register_nvs_regions(void)
779{
780 int i;
781
782 for (i = 0; i < e820_table->nr_entries; i++) {
783 struct e820_entry *entry = &e820_table->entries[i];
784
785 if (entry->type == E820_TYPE_NVS)
786 acpi_nvs_register(start: entry->addr, size: entry->size);
787 }
788
789 return 0;
790}
791core_initcall(e820__register_nvs_regions);
792#endif
793
794/*
795 * Allocate the requested number of bytes with the requested alignment
796 * and return (the physical address) to the caller. Also register this
797 * range in the 'kexec' E820 table as a reserved range.
798 *
799 * This allows kexec to fake a new mptable, as if it came from the real
800 * system.
801 */
802u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
803{
804 u64 addr;
805
806 addr = memblock_phys_alloc(size, align);
807 if (addr) {
808 e820__range_update_table(t: e820_table_kexec, start: addr, size, old_type: E820_TYPE_RAM, new_type: E820_TYPE_RESERVED);
809 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
810 e820__update_table_kexec();
811 }
812
813 return addr;
814}
815
816#ifdef CONFIG_X86_32
817# ifdef CONFIG_X86_PAE
818# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
819# else
820# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
821# endif
822#else /* CONFIG_X86_32 */
823# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
824#endif
825
826/*
827 * Find the highest page frame number we have available
828 */
829static unsigned long __init e820__end_ram_pfn(unsigned long limit_pfn)
830{
831 int i;
832 unsigned long last_pfn = 0;
833 unsigned long max_arch_pfn = MAX_ARCH_PFN;
834
835 for (i = 0; i < e820_table->nr_entries; i++) {
836 struct e820_entry *entry = &e820_table->entries[i];
837 unsigned long start_pfn;
838 unsigned long end_pfn;
839
840 if (entry->type != E820_TYPE_RAM &&
841 entry->type != E820_TYPE_ACPI)
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_ram_pfn(MAX_ARCH_PFN);
868}
869
870unsigned long __init e820__end_of_low_ram_pfn(void)
871{
872 return e820__end_ram_pfn(limit_pfn: 1UL << (32 - PAGE_SHIFT));
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 * Called after parse_early_param(), after early parameters (such as mem=)
993 * have been processed, in which case we already have an E820 table filled in
994 * via the parameter callback function(s), but it's not sorted and printed yet:
995 */
996void __init e820__finish_early_params(void)
997{
998 if (userdef) {
999 if (e820__update_table(table: e820_table) < 0)
1000 early_panic(msg: "Invalid user supplied memory map");
1001
1002 pr_info("user-defined physical RAM map:\n");
1003 e820__print_table(who: "user");
1004 }
1005}
1006
1007static const char *__init e820_type_to_string(struct e820_entry *entry)
1008{
1009 switch (entry->type) {
1010 case E820_TYPE_RAM: return "System RAM";
1011 case E820_TYPE_ACPI: return "ACPI Tables";
1012 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1013 case E820_TYPE_UNUSABLE: return "Unusable memory";
1014 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1015 case E820_TYPE_PMEM: return "Persistent Memory";
1016 case E820_TYPE_RESERVED: return "Reserved";
1017 case E820_TYPE_SOFT_RESERVED: return "Soft Reserved";
1018 default: return "Unknown E820 type";
1019 }
1020}
1021
1022static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1023{
1024 switch (entry->type) {
1025 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1026 case E820_TYPE_ACPI: /* Fall-through: */
1027 case E820_TYPE_NVS: /* Fall-through: */
1028 case E820_TYPE_UNUSABLE: /* Fall-through: */
1029 case E820_TYPE_PRAM: /* Fall-through: */
1030 case E820_TYPE_PMEM: /* Fall-through: */
1031 case E820_TYPE_RESERVED: /* Fall-through: */
1032 case E820_TYPE_SOFT_RESERVED: /* Fall-through: */
1033 default: return IORESOURCE_MEM;
1034 }
1035}
1036
1037static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1038{
1039 switch (entry->type) {
1040 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1041 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1042 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1043 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1044 case E820_TYPE_RESERVED: return IORES_DESC_RESERVED;
1045 case E820_TYPE_SOFT_RESERVED: return IORES_DESC_SOFT_RESERVED;
1046 case E820_TYPE_RAM: /* Fall-through: */
1047 case E820_TYPE_UNUSABLE: /* Fall-through: */
1048 default: return IORES_DESC_NONE;
1049 }
1050}
1051
1052static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1053{
1054 /* this is the legacy bios/dos rom-shadow + mmio region */
1055 if (res->start < (1ULL<<20))
1056 return true;
1057
1058 /*
1059 * Treat persistent memory and other special memory ranges like
1060 * device memory, i.e. reserve it for exclusive use of a driver
1061 */
1062 switch (type) {
1063 case E820_TYPE_RESERVED:
1064 case E820_TYPE_SOFT_RESERVED:
1065 case E820_TYPE_PRAM:
1066 case E820_TYPE_PMEM:
1067 return false;
1068 case E820_TYPE_RAM:
1069 case E820_TYPE_ACPI:
1070 case E820_TYPE_NVS:
1071 case E820_TYPE_UNUSABLE:
1072 default:
1073 return true;
1074 }
1075}
1076
1077/*
1078 * Mark E820 reserved areas as busy for the resource manager:
1079 */
1080
1081static struct resource __initdata *e820_res;
1082
1083void __init e820__reserve_resources(void)
1084{
1085 int i;
1086 struct resource *res;
1087 u64 end;
1088
1089 res = memblock_alloc_or_panic(sizeof(*res) * e820_table->nr_entries,
1090 SMP_CACHE_BYTES);
1091 e820_res = res;
1092
1093 for (i = 0; i < e820_table->nr_entries; i++) {
1094 struct e820_entry *entry = e820_table->entries + i;
1095
1096 end = entry->addr + entry->size - 1;
1097 if (end != (resource_size_t)end) {
1098 res++;
1099 continue;
1100 }
1101 res->start = entry->addr;
1102 res->end = end;
1103 res->name = e820_type_to_string(entry);
1104 res->flags = e820_type_to_iomem_type(entry);
1105 res->desc = e820_type_to_iores_desc(entry);
1106
1107 /*
1108 * Don't register the region that could be conflicted with
1109 * PCI device BAR resources and insert them later in
1110 * pcibios_resource_survey():
1111 */
1112 if (do_mark_busy(type: entry->type, res)) {
1113 res->flags |= IORESOURCE_BUSY;
1114 insert_resource(parent: &iomem_resource, new: res);
1115 }
1116 res++;
1117 }
1118
1119 /* Expose the kexec e820 table to the sysfs. */
1120 for (i = 0; i < e820_table_kexec->nr_entries; i++) {
1121 struct e820_entry *entry = e820_table_kexec->entries + i;
1122
1123 firmware_map_add_early(start: entry->addr, end: entry->addr + entry->size, type: e820_type_to_string(entry));
1124 }
1125}
1126
1127/*
1128 * How much should we pad the end of RAM, depending on where it is?
1129 */
1130static unsigned long __init ram_alignment(resource_size_t pos)
1131{
1132 unsigned long mb = pos >> 20;
1133
1134 /* To 64kB in the first megabyte */
1135 if (!mb)
1136 return 64*1024;
1137
1138 /* To 1MB in the first 16MB */
1139 if (mb < 16)
1140 return 1024*1024;
1141
1142 /* To 64MB for anything above that */
1143 return 64*1024*1024;
1144}
1145
1146#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1147
1148void __init e820__reserve_resources_late(void)
1149{
1150 int i;
1151 struct resource *res;
1152
1153 res = e820_res;
1154 for (i = 0; i < e820_table->nr_entries; i++) {
1155 if (!res->parent && res->end)
1156 insert_resource_expand_to_fit(root: &iomem_resource, new: res);
1157 res++;
1158 }
1159
1160 /*
1161 * Try to bump up RAM regions to reasonable boundaries, to
1162 * avoid stolen RAM:
1163 */
1164 for (i = 0; i < e820_table->nr_entries; i++) {
1165 struct e820_entry *entry = &e820_table->entries[i];
1166 u64 start, end;
1167
1168 if (entry->type != E820_TYPE_RAM)
1169 continue;
1170
1171 start = entry->addr + entry->size;
1172 end = round_up(start, ram_alignment(start)) - 1;
1173 if (end > MAX_RESOURCE_SIZE)
1174 end = MAX_RESOURCE_SIZE;
1175 if (start >= end)
1176 continue;
1177
1178 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1179 reserve_region_with_split(root: &iomem_resource, start, end, name: "RAM buffer");
1180 }
1181}
1182
1183/*
1184 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1185 */
1186char *__init e820__memory_setup_default(void)
1187{
1188 char *who = "BIOS-e820";
1189
1190 /*
1191 * Try to copy the BIOS-supplied E820-map.
1192 *
1193 * Otherwise fake a memory map; one section from 0k->640k,
1194 * the next section from 1mb->appropriate_mem_k
1195 */
1196 if (append_e820_table(entries: boot_params.e820_table, nr_entries: boot_params.e820_entries) < 0) {
1197 u64 mem_size;
1198
1199 /* Compare results from other methods and take the one that gives more RAM: */
1200 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1201 mem_size = boot_params.screen_info.ext_mem_k;
1202 who = "BIOS-88";
1203 } else {
1204 mem_size = boot_params.alt_mem_k;
1205 who = "BIOS-e801";
1206 }
1207
1208 e820_table->nr_entries = 0;
1209 e820__range_add(start: 0, LOWMEMSIZE(), type: E820_TYPE_RAM);
1210 e820__range_add(HIGH_MEMORY, size: mem_size << 10, type: E820_TYPE_RAM);
1211 }
1212
1213 /* We just appended a lot of ranges, sanitize the table: */
1214 e820__update_table(table: e820_table);
1215
1216 return who;
1217}
1218
1219/*
1220 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1221 * E820 map - with an optional platform quirk available for virtual platforms
1222 * to override this method of boot environment processing:
1223 */
1224void __init e820__memory_setup(void)
1225{
1226 char *who;
1227
1228 /* This is a firmware interface ABI - make sure we don't break it: */
1229 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1230
1231 who = x86_init.resources.memory_setup();
1232
1233 memcpy(to: e820_table_kexec, from: e820_table, len: sizeof(*e820_table_kexec));
1234 memcpy(to: e820_table_firmware, from: e820_table, len: sizeof(*e820_table_firmware));
1235
1236 pr_info("BIOS-provided physical RAM map:\n");
1237 e820__print_table(who);
1238}
1239
1240void __init e820__memblock_setup(void)
1241{
1242 int i;
1243 u64 end;
1244
1245#ifdef CONFIG_MEMORY_HOTPLUG
1246 /*
1247 * Memory used by the kernel cannot be hot-removed because Linux
1248 * cannot migrate the kernel pages. When memory hotplug is
1249 * enabled, we should prevent memblock from allocating memory
1250 * for the kernel.
1251 *
1252 * ACPI SRAT records all hotpluggable memory ranges. But before
1253 * SRAT is parsed, we don't know about it.
1254 *
1255 * The kernel image is loaded into memory at very early time. We
1256 * cannot prevent this anyway. So on NUMA system, we set any
1257 * node the kernel resides in as un-hotpluggable.
1258 *
1259 * Since on modern servers, one node could have double-digit
1260 * gigabytes memory, we can assume the memory around the kernel
1261 * image is also un-hotpluggable. So before SRAT is parsed, just
1262 * allocate memory near the kernel image to try the best to keep
1263 * the kernel away from hotpluggable memory.
1264 */
1265 if (movable_node_is_enabled())
1266 memblock_set_bottom_up(true);
1267#endif
1268
1269 /*
1270 * At this point only the first megabyte is mapped for sure, the
1271 * rest of the memory cannot be used for memblock resizing
1272 */
1273 memblock_set_current_limit(ISA_END_ADDRESS);
1274
1275 /*
1276 * The bootstrap memblock region count maximum is 128 entries
1277 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1278 * than that - so allow memblock resizing.
1279 *
1280 * This is safe, because this call happens pretty late during x86 setup,
1281 * so we know about reserved memory regions already. (This is important
1282 * so that memblock resizing does no stomp over reserved areas.)
1283 */
1284 memblock_allow_resize();
1285
1286 for (i = 0; i < e820_table->nr_entries; i++) {
1287 struct e820_entry *entry = &e820_table->entries[i];
1288
1289 end = entry->addr + entry->size;
1290 if (end != (resource_size_t)end)
1291 continue;
1292
1293 if (entry->type == E820_TYPE_SOFT_RESERVED)
1294 memblock_reserve(base: entry->addr, size: entry->size);
1295
1296 if (entry->type != E820_TYPE_RAM)
1297 continue;
1298
1299 memblock_add(base: entry->addr, size: entry->size);
1300 }
1301
1302 /*
1303 * At this point memblock is only allowed to allocate from memory
1304 * below 1M (aka ISA_END_ADDRESS) up until direct map is completely set
1305 * up in init_mem_mapping().
1306 *
1307 * KHO kernels are special and use only scratch memory for memblock
1308 * allocations, but memory below 1M is ignored by kernel after early
1309 * boot and cannot be naturally marked as scratch.
1310 *
1311 * To allow allocation of the real-mode trampoline and a few (if any)
1312 * other very early allocations from below 1M forcibly mark the memory
1313 * below 1M as scratch.
1314 *
1315 * After real mode trampoline is allocated, we clear that scratch
1316 * marking.
1317 */
1318 memblock_mark_kho_scratch(base: 0, SZ_1M);
1319
1320 /*
1321 * 32-bit systems are limited to 4BG of memory even with HIGHMEM and
1322 * to even less without it.
1323 * Discard memory after max_pfn - the actual limit detected at runtime.
1324 */
1325 if (IS_ENABLED(CONFIG_X86_32))
1326 memblock_remove(PFN_PHYS(max_pfn), size: -1);
1327
1328 /* Throw away partial pages: */
1329 memblock_trim_memory(PAGE_SIZE);
1330
1331 memblock_dump_all();
1332}
1333