Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Low level x86 E820 memory map handling functions.
4 *
5 * The firmware and bootloader passes us the "E820 table", which is the primary
6 * physical memory layout description available about x86 systems.
7 *
8 * The kernel takes the E820 memory layout and optionally modifies it with
9 * quirks and other tweaks, and feeds that into the generic Linux memory
10 * allocation code routines via a platform independent interface (memblock, etc.).
11 */
12#include <linux/crash_dump.h>
13#include <linux/memblock.h>
14#include <linux/suspend.h>
15#include <linux/acpi.h>
16#include <linux/firmware-map.h>
17#include <linux/sort.h>
18#include <linux/memory_hotplug.h>
19
20#include <asm/e820/api.h>
21#include <asm/setup.h>
22
23/*
24 * We organize the E820 table into three main data structures:
25 *
26 * - 'e820_table_firmware': the original firmware version passed to us by the
27 * bootloader - not modified by the kernel. It is composed of two parts:
28 * the first 128 E820 memory entries in boot_params.e820_table and the remaining
29 * (if any) entries of the SETUP_E820_EXT nodes. We use this to:
30 *
31 * - inform the user about the firmware's notion of memory layout
32 * via /sys/firmware/memmap
33 *
34 * - the hibernation code uses it to generate a kernel-independent MD5
35 * fingerprint of the physical memory layout of a system.
36 *
37 * - 'e820_table_kexec': a slightly modified (by the kernel) firmware version
38 * passed to us by the bootloader - the major difference between
39 * e820_table_firmware[] and this one is that, the latter marks the setup_data
40 * list created by the EFI boot stub as reserved, so that kexec can reuse the
41 * setup_data information in the second kernel. Besides, e820_table_kexec[]
42 * might also be modified by the kexec itself to fake a mptable.
43 * We use this to:
44 *
45 * - kexec, which is a bootloader in disguise, uses the original E820
46 * layout to pass to the kexec-ed kernel. This way the original kernel
47 * can have a restricted E820 map while the kexec()-ed kexec-kernel
48 * can have access to full memory - etc.
49 *
50 * - 'e820_table': this is the main E820 table that is massaged by the
51 * low level x86 platform code, or modified by boot parameters, before
52 * passed on to higher level MM layers.
53 *
54 * Once the E820 map has been converted to the standard Linux memory layout
55 * information its role stops - modifying it has no effect and does not get
56 * re-propagated. So itsmain role is a temporary bootstrap storage of firmware
57 * specific memory layout data during early bootup.
58 */
59static struct e820_table e820_table_init __initdata;
60static struct e820_table e820_table_kexec_init __initdata;
61static struct e820_table e820_table_firmware_init __initdata;
62
63struct e820_table *e820_table __refdata = &e820_table_init;
64struct e820_table *e820_table_kexec __refdata = &e820_table_kexec_init;
65struct e820_table *e820_table_firmware __refdata = &e820_table_firmware_init;
66
67/* For PCI or other memory-mapped resources */
68unsigned long pci_mem_start = 0xaeedbabe;
69#ifdef CONFIG_PCI
70EXPORT_SYMBOL(pci_mem_start);
71#endif
72
73/*
74 * This function checks if any part of the range <start,end> is mapped
75 * with type.
76 */
77static bool _e820__mapped_any(struct e820_table *table,
78 u64 start, u64 end, enum e820_type type)
79{
80 int i;
81
82 for (i = 0; i < table->nr_entries; i++) {
83 struct e820_entry *entry = &table->entries[i];
84
85 if (type && entry->type != type)
86 continue;
87 if (entry->addr >= end || entry->addr + entry->size <= start)
88 continue;
89 return true;
90 }
91 return false;
92}
93
94bool e820__mapped_raw_any(u64 start, u64 end, enum e820_type type)
95{
96 return _e820__mapped_any(e820_table_firmware, start, end, type);
97}
98EXPORT_SYMBOL_GPL(e820__mapped_raw_any);
99
100bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
101{
102 return _e820__mapped_any(e820_table, start, end, type);
103}
104EXPORT_SYMBOL_GPL(e820__mapped_any);
105
106/*
107 * This function checks if the entire <start,end> range is mapped with 'type'.
108 *
109 * Note: this function only works correctly once the E820 table is sorted and
110 * not-overlapping (at least for the range specified), which is the case normally.
111 */
112static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
113 enum e820_type type)
114{
115 int i;
116
117 for (i = 0; i < e820_table->nr_entries; i++) {
118 struct e820_entry *entry = &e820_table->entries[i];
119
120 if (type && entry->type != type)
121 continue;
122
123 /* Is the region (part) in overlap with the current region? */
124 if (entry->addr >= end || entry->addr + entry->size <= start)
125 continue;
126
127 /*
128 * If the region is at the beginning of <start,end> we move
129 * 'start' to the end of the region since it's ok until there
130 */
131 if (entry->addr <= start)
132 start = entry->addr + entry->size;
133
134 /*
135 * If 'start' is now at or beyond 'end', we're done, full
136 * coverage of the desired range exists:
137 */
138 if (start >= end)
139 return entry;
140 }
141
142 return NULL;
143}
144
145/*
146 * This function checks if the entire range <start,end> is mapped with type.
147 */
148bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
149{
150 return __e820__mapped_all(start, end, type);
151}
152
153/*
154 * This function returns the type associated with the range <start,end>.
155 */
156int e820__get_entry_type(u64 start, u64 end)
157{
158 struct e820_entry *entry = __e820__mapped_all(start, end, 0);
159
160 return entry ? entry->type : -EINVAL;
161}
162
163/*
164 * Add a memory region to the kernel E820 map.
165 */
166static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
167{
168 int x = table->nr_entries;
169
170 if (x >= ARRAY_SIZE(table->entries)) {
171 pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
172 start, start + size - 1);
173 return;
174 }
175
176 table->entries[x].addr = start;
177 table->entries[x].size = size;
178 table->entries[x].type = type;
179 table->nr_entries++;
180}
181
182void __init e820__range_add(u64 start, u64 size, enum e820_type type)
183{
184 __e820__range_add(e820_table, start, size, type);
185}
186
187static void __init e820_print_type(enum e820_type type)
188{
189 switch (type) {
190 case E820_TYPE_RAM: /* Fall through: */
191 case E820_TYPE_RESERVED_KERN: pr_cont("usable"); break;
192 case E820_TYPE_RESERVED: pr_cont("reserved"); break;
193 case E820_TYPE_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(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
307int __init e820__update_table(struct e820_table *table)
308{
309 struct e820_entry *entries = table->entries;
310 u32 max_nr_entries = ARRAY_SIZE(table->entries);
311 enum e820_type current_type, last_type;
312 unsigned long long last_addr;
313 u32 new_nr_entries, overlap_entries;
314 u32 i, chg_idx, chg_nr;
315
316 /* If there's only one memory region, don't bother: */
317 if (table->nr_entries < 2)
318 return -1;
319
320 BUG_ON(table->nr_entries > max_nr_entries);
321
322 /* Bail out if we find any unreasonable addresses in the map: */
323 for (i = 0; i < table->nr_entries; i++) {
324 if (entries[i].addr + entries[i].size < entries[i].addr)
325 return -1;
326 }
327
328 /* Create pointers for initial change-point information (for sorting): */
329 for (i = 0; i < 2 * table->nr_entries; i++)
330 change_point[i] = &change_point_list[i];
331
332 /*
333 * Record all known change-points (starting and ending addresses),
334 * omitting empty memory regions:
335 */
336 chg_idx = 0;
337 for (i = 0; i < table->nr_entries; i++) {
338 if (entries[i].size != 0) {
339 change_point[chg_idx]->addr = entries[i].addr;
340 change_point[chg_idx++]->entry = &entries[i];
341 change_point[chg_idx]->addr = entries[i].addr + entries[i].size;
342 change_point[chg_idx++]->entry = &entries[i];
343 }
344 }
345 chg_nr = chg_idx;
346
347 /* Sort change-point list by memory addresses (low -> high): */
348 sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
349
350 /* Create a new memory map, removing overlaps: */
351 overlap_entries = 0; /* Number of entries in the overlap table */
352 new_nr_entries = 0; /* Index for creating new map entries */
353 last_type = 0; /* Start with undefined memory type */
354 last_addr = 0; /* Start with 0 as last starting address */
355
356 /* Loop through change-points, determining effect on the new map: */
357 for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
358 /* Keep track of all overlapping entries */
359 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
360 /* Add map entry to overlap list (> 1 entry implies an overlap) */
361 overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
362 } else {
363 /* Remove entry from list (order independent, so swap with last): */
364 for (i = 0; i < overlap_entries; i++) {
365 if (overlap_list[i] == change_point[chg_idx]->entry)
366 overlap_list[i] = overlap_list[overlap_entries-1];
367 }
368 overlap_entries--;
369 }
370 /*
371 * If there are overlapping entries, decide which
372 * "type" to use (larger value takes precedence --
373 * 1=usable, 2,3,4,4+=unusable)
374 */
375 current_type = 0;
376 for (i = 0; i < overlap_entries; i++) {
377 if (overlap_list[i]->type > current_type)
378 current_type = overlap_list[i]->type;
379 }
380
381 /* Continue building up new map based on this information: */
382 if (current_type != last_type || current_type == E820_TYPE_PRAM) {
383 if (last_type != 0) {
384 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
385 /* Move forward only if the new size was non-zero: */
386 if (new_entries[new_nr_entries].size != 0)
387 /* No more space left for new entries? */
388 if (++new_nr_entries >= max_nr_entries)
389 break;
390 }
391 if (current_type != 0) {
392 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
393 new_entries[new_nr_entries].type = current_type;
394 last_addr = change_point[chg_idx]->addr;
395 }
396 last_type = current_type;
397 }
398 }
399
400 /* Copy the new entries into the original location: */
401 memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
402 table->nr_entries = new_nr_entries;
403
404 return 0;
405}
406
407static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
408{
409 struct boot_e820_entry *entry = entries;
410
411 while (nr_entries) {
412 u64 start = entry->addr;
413 u64 size = entry->size;
414 u64 end = start + size - 1;
415 u32 type = entry->type;
416
417 /* Ignore the entry on 64-bit overflow: */
418 if (start > end && likely(size))
419 return -1;
420
421 e820__range_add(start, size, type);
422
423 entry++;
424 nr_entries--;
425 }
426 return 0;
427}
428
429/*
430 * Copy the BIOS E820 map into a safe place.
431 *
432 * Sanity-check it while we're at it..
433 *
434 * If we're lucky and live on a modern system, the setup code
435 * will have given us a memory map that we can use to properly
436 * set up memory. If we aren't, we'll fake a memory map.
437 */
438static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
439{
440 /* Only one memory region (or negative)? Ignore it */
441 if (nr_entries < 2)
442 return -1;
443
444 return __append_e820_table(entries, nr_entries);
445}
446
447static u64 __init
448__e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
449{
450 u64 end;
451 unsigned int i;
452 u64 real_updated_size = 0;
453
454 BUG_ON(old_type == new_type);
455
456 if (size > (ULLONG_MAX - start))
457 size = ULLONG_MAX - start;
458
459 end = start + size;
460 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
461 e820_print_type(old_type);
462 pr_cont(" ==> ");
463 e820_print_type(new_type);
464 pr_cont("\n");
465
466 for (i = 0; i < table->nr_entries; i++) {
467 struct e820_entry *entry = &table->entries[i];
468 u64 final_start, final_end;
469 u64 entry_end;
470
471 if (entry->type != old_type)
472 continue;
473
474 entry_end = entry->addr + entry->size;
475
476 /* Completely covered by new range? */
477 if (entry->addr >= start && entry_end <= end) {
478 entry->type = new_type;
479 real_updated_size += entry->size;
480 continue;
481 }
482
483 /* New range is completely covered? */
484 if (entry->addr < start && entry_end > end) {
485 __e820__range_add(table, start, size, new_type);
486 __e820__range_add(table, end, entry_end - end, entry->type);
487 entry->size = start - entry->addr;
488 real_updated_size += size;
489 continue;
490 }
491
492 /* Partially covered: */
493 final_start = max(start, entry->addr);
494 final_end = min(end, entry_end);
495 if (final_start >= final_end)
496 continue;
497
498 __e820__range_add(table, final_start, final_end - final_start, new_type);
499
500 real_updated_size += final_end - final_start;
501
502 /*
503 * Left range could be head or tail, so need to update
504 * its size first:
505 */
506 entry->size -= final_end - final_start;
507 if (entry->addr < final_start)
508 continue;
509
510 entry->addr = final_end;
511 }
512 return real_updated_size;
513}
514
515u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
516{
517 return __e820__range_update(e820_table, start, size, old_type, new_type);
518}
519
520static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
521{
522 return __e820__range_update(e820_table_kexec, start, size, old_type, new_type);
523}
524
525/* Remove a range of memory from the E820 table: */
526u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
527{
528 int i;
529 u64 end;
530 u64 real_removed_size = 0;
531
532 if (size > (ULLONG_MAX - start))
533 size = ULLONG_MAX - start;
534
535 end = start + size;
536 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
537 if (check_type)
538 e820_print_type(old_type);
539 pr_cont("\n");
540
541 for (i = 0; i < e820_table->nr_entries; i++) {
542 struct e820_entry *entry = &e820_table->entries[i];
543 u64 final_start, final_end;
544 u64 entry_end;
545
546 if (check_type && entry->type != old_type)
547 continue;
548
549 entry_end = entry->addr + entry->size;
550
551 /* Completely covered? */
552 if (entry->addr >= start && entry_end <= end) {
553 real_removed_size += entry->size;
554 memset(entry, 0, sizeof(*entry));
555 continue;
556 }
557
558 /* Is the new range completely covered? */
559 if (entry->addr < start && entry_end > end) {
560 e820__range_add(end, entry_end - end, entry->type);
561 entry->size = start - entry->addr;
562 real_removed_size += size;
563 continue;
564 }
565
566 /* Partially covered: */
567 final_start = max(start, entry->addr);
568 final_end = min(end, entry_end);
569 if (final_start >= final_end)
570 continue;
571
572 real_removed_size += final_end - final_start;
573
574 /*
575 * Left range could be head or tail, so need to update
576 * the size first:
577 */
578 entry->size -= final_end - final_start;
579 if (entry->addr < final_start)
580 continue;
581
582 entry->addr = final_end;
583 }
584 return real_removed_size;
585}
586
587void __init e820__update_table_print(void)
588{
589 if (e820__update_table(e820_table))
590 return;
591
592 pr_info("modified physical RAM map:\n");
593 e820__print_table("modified");
594}
595
596static void __init e820__update_table_kexec(void)
597{
598 e820__update_table(e820_table_kexec);
599}
600
601#define MAX_GAP_END 0x100000000ull
602
603/*
604 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
605 */
606static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
607{
608 unsigned long long last = MAX_GAP_END;
609 int i = e820_table->nr_entries;
610 int found = 0;
611
612 while (--i >= 0) {
613 unsigned long long start = e820_table->entries[i].addr;
614 unsigned long long end = start + e820_table->entries[i].size;
615
616 /*
617 * Since "last" is at most 4GB, we know we'll
618 * fit in 32 bits if this condition is true:
619 */
620 if (last > end) {
621 unsigned long gap = last - end;
622
623 if (gap >= *gapsize) {
624 *gapsize = gap;
625 *gapstart = end;
626 found = 1;
627 }
628 }
629 if (start < last)
630 last = start;
631 }
632 return found;
633}
634
635/*
636 * Search for the biggest gap in the low 32 bits of the E820
637 * memory space. We pass this space to the PCI subsystem, so
638 * that it can assign MMIO resources for hotplug or
639 * unconfigured devices in.
640 *
641 * Hopefully the BIOS let enough space left.
642 */
643__init void e820__setup_pci_gap(void)
644{
645 unsigned long gapstart, gapsize;
646 int found;
647
648 gapsize = 0x400000;
649 found = e820_search_gap(&gapstart, &gapsize);
650
651 if (!found) {
652#ifdef CONFIG_X86_64
653 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
654 pr_err("Cannot find an available gap in the 32-bit address range\n");
655 pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
656#else
657 gapstart = 0x10000000;
658#endif
659 }
660
661 /*
662 * e820__reserve_resources_late() protects stolen RAM already:
663 */
664 pci_mem_start = gapstart;
665
666 pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
667 gapstart, gapstart + gapsize - 1);
668}
669
670/*
671 * Called late during init, in free_initmem().
672 *
673 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
674 *
675 * Copy them to a (usually much smaller) dynamically allocated area that is
676 * sized precisely after the number of e820 entries.
677 *
678 * This is done after we've performed all the fixes and tweaks to the tables.
679 * All functions which modify them are __init functions, which won't exist
680 * after free_initmem().
681 */
682__init void e820__reallocate_tables(void)
683{
684 struct e820_table *n;
685 int size;
686
687 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
688 n = kmemdup(e820_table, size, GFP_KERNEL);
689 BUG_ON(!n);
690 e820_table = n;
691
692 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
693 n = kmemdup(e820_table_kexec, size, GFP_KERNEL);
694 BUG_ON(!n);
695 e820_table_kexec = n;
696
697 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
698 n = kmemdup(e820_table_firmware, size, GFP_KERNEL);
699 BUG_ON(!n);
700 e820_table_firmware = n;
701}
702
703/*
704 * Because of the small fixed size of struct boot_params, only the first
705 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
706 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
707 * struct setup_data, which is parsed here.
708 */
709void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
710{
711 int entries;
712 struct boot_e820_entry *extmap;
713 struct setup_data *sdata;
714
715 sdata = early_memremap(phys_addr, data_len);
716 entries = sdata->len / sizeof(*extmap);
717 extmap = (struct boot_e820_entry *)(sdata->data);
718
719 __append_e820_table(extmap, entries);
720 e820__update_table(e820_table);
721
722 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
723 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
724
725 early_memunmap(sdata, data_len);
726 pr_info("extended physical RAM map:\n");
727 e820__print_table("extended");
728}
729
730/*
731 * Find the ranges of physical addresses that do not correspond to
732 * E820 RAM areas and register the corresponding pages as 'nosave' for
733 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
734 *
735 * This function requires the E820 map to be sorted and without any
736 * overlapping entries.
737 */
738void __init e820__register_nosave_regions(unsigned long limit_pfn)
739{
740 int i;
741 unsigned long pfn = 0;
742
743 for (i = 0; i < e820_table->nr_entries; i++) {
744 struct e820_entry *entry = &e820_table->entries[i];
745
746 if (pfn < PFN_UP(entry->addr))
747 register_nosave_region(pfn, PFN_UP(entry->addr));
748
749 pfn = PFN_DOWN(entry->addr + entry->size);
750
751 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
752 register_nosave_region(PFN_UP(entry->addr), pfn);
753
754 if (pfn >= limit_pfn)
755 break;
756 }
757}
758
759#ifdef CONFIG_ACPI
760/*
761 * Register ACPI NVS memory regions, so that we can save/restore them during
762 * hibernation and the subsequent resume:
763 */
764static int __init e820__register_nvs_regions(void)
765{
766 int i;
767
768 for (i = 0; i < e820_table->nr_entries; i++) {
769 struct e820_entry *entry = &e820_table->entries[i];
770
771 if (entry->type == E820_TYPE_NVS)
772 acpi_nvs_register(entry->addr, entry->size);
773 }
774
775 return 0;
776}
777core_initcall(e820__register_nvs_regions);
778#endif
779
780/*
781 * Allocate the requested number of bytes with the requsted alignment
782 * and return (the physical address) to the caller. Also register this
783 * range in the 'kexec' E820 table as a reserved range.
784 *
785 * This allows kexec to fake a new mptable, as if it came from the real
786 * system.
787 */
788u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
789{
790 u64 addr;
791
792 addr = memblock_phys_alloc(size, align);
793 if (addr) {
794 e820__range_update_kexec(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
795 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
796 e820__update_table_kexec();
797 }
798
799 return addr;
800}
801
802#ifdef CONFIG_X86_32
803# ifdef CONFIG_X86_PAE
804# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
805# else
806# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
807# endif
808#else /* CONFIG_X86_32 */
809# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
810#endif
811
812/*
813 * Find the highest page frame number we have available
814 */
815static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
816{
817 int i;
818 unsigned long last_pfn = 0;
819 unsigned long max_arch_pfn = MAX_ARCH_PFN;
820
821 for (i = 0; i < e820_table->nr_entries; i++) {
822 struct e820_entry *entry = &e820_table->entries[i];
823 unsigned long start_pfn;
824 unsigned long end_pfn;
825
826 if (entry->type != type)
827 continue;
828
829 start_pfn = entry->addr >> PAGE_SHIFT;
830 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
831
832 if (start_pfn >= limit_pfn)
833 continue;
834 if (end_pfn > limit_pfn) {
835 last_pfn = limit_pfn;
836 break;
837 }
838 if (end_pfn > last_pfn)
839 last_pfn = end_pfn;
840 }
841
842 if (last_pfn > max_arch_pfn)
843 last_pfn = max_arch_pfn;
844
845 pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
846 last_pfn, max_arch_pfn);
847 return last_pfn;
848}
849
850unsigned long __init e820__end_of_ram_pfn(void)
851{
852 return e820_end_pfn(MAX_ARCH_PFN, E820_TYPE_RAM);
853}
854
855unsigned long __init e820__end_of_low_ram_pfn(void)
856{
857 return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
858}
859
860static void __init early_panic(char *msg)
861{
862 early_printk(msg);
863 panic(msg);
864}
865
866static int userdef __initdata;
867
868/* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
869static int __init parse_memopt(char *p)
870{
871 u64 mem_size;
872
873 if (!p)
874 return -EINVAL;
875
876 if (!strcmp(p, "nopentium")) {
877#ifdef CONFIG_X86_32
878 setup_clear_cpu_cap(X86_FEATURE_PSE);
879 return 0;
880#else
881 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
882 return -EINVAL;
883#endif
884 }
885
886 userdef = 1;
887 mem_size = memparse(p, &p);
888
889 /* Don't remove all memory when getting "mem={invalid}" parameter: */
890 if (mem_size == 0)
891 return -EINVAL;
892
893 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
894
895#ifdef CONFIG_MEMORY_HOTPLUG
896 max_mem_size = mem_size;
897#endif
898
899 return 0;
900}
901early_param("mem", parse_memopt);
902
903static int __init parse_memmap_one(char *p)
904{
905 char *oldp;
906 u64 start_at, mem_size;
907
908 if (!p)
909 return -EINVAL;
910
911 if (!strncmp(p, "exactmap", 8)) {
912#ifdef CONFIG_CRASH_DUMP
913 /*
914 * If we are doing a crash dump, we still need to know
915 * the real memory size before the original memory map is
916 * reset.
917 */
918 saved_max_pfn = e820__end_of_ram_pfn();
919#endif
920 e820_table->nr_entries = 0;
921 userdef = 1;
922 return 0;
923 }
924
925 oldp = p;
926 mem_size = memparse(p, &p);
927 if (p == oldp)
928 return -EINVAL;
929
930 userdef = 1;
931 if (*p == '@') {
932 start_at = memparse(p+1, &p);
933 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
934 } else if (*p == '#') {
935 start_at = memparse(p+1, &p);
936 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
937 } else if (*p == '$') {
938 start_at = memparse(p+1, &p);
939 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
940 } else if (*p == '!') {
941 start_at = memparse(p+1, &p);
942 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
943 } else if (*p == '%') {
944 enum e820_type from = 0, to = 0;
945
946 start_at = memparse(p + 1, &p);
947 if (*p == '-')
948 from = simple_strtoull(p + 1, &p, 0);
949 if (*p == '+')
950 to = simple_strtoull(p + 1, &p, 0);
951 if (*p != '\0')
952 return -EINVAL;
953 if (from && to)
954 e820__range_update(start_at, mem_size, from, to);
955 else if (to)
956 e820__range_add(start_at, mem_size, to);
957 else if (from)
958 e820__range_remove(start_at, mem_size, from, 1);
959 else
960 e820__range_remove(start_at, mem_size, 0, 0);
961 } else {
962 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
963 }
964
965 return *p == '\0' ? 0 : -EINVAL;
966}
967
968static int __init parse_memmap_opt(char *str)
969{
970 while (str) {
971 char *k = strchr(str, ',');
972
973 if (k)
974 *k++ = 0;
975
976 parse_memmap_one(str);
977 str = k;
978 }
979
980 return 0;
981}
982early_param("memmap", parse_memmap_opt);
983
984/*
985 * Reserve all entries from the bootloader's extensible data nodes list,
986 * because if present we are going to use it later on to fetch e820
987 * entries from it:
988 */
989void __init e820__reserve_setup_data(void)
990{
991 struct setup_data *data;
992 u64 pa_data;
993
994 pa_data = boot_params.hdr.setup_data;
995 if (!pa_data)
996 return;
997
998 while (pa_data) {
999 data = early_memremap(pa_data, sizeof(*data));
1000 e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1001 e820__range_update_kexec(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1002 pa_data = data->next;
1003 early_memunmap(data, sizeof(*data));
1004 }
1005
1006 e820__update_table(e820_table);
1007 e820__update_table(e820_table_kexec);
1008
1009 pr_info("extended physical RAM map:\n");
1010 e820__print_table("reserve setup_data");
1011}
1012
1013/*
1014 * Called after parse_early_param(), after early parameters (such as mem=)
1015 * have been processed, in which case we already have an E820 table filled in
1016 * via the parameter callback function(s), but it's not sorted and printed yet:
1017 */
1018void __init e820__finish_early_params(void)
1019{
1020 if (userdef) {
1021 if (e820__update_table(e820_table) < 0)
1022 early_panic("Invalid user supplied memory map");
1023
1024 pr_info("user-defined physical RAM map:\n");
1025 e820__print_table("user");
1026 }
1027}
1028
1029static const char *__init e820_type_to_string(struct e820_entry *entry)
1030{
1031 switch (entry->type) {
1032 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1033 case E820_TYPE_RAM: return "System RAM";
1034 case E820_TYPE_ACPI: return "ACPI Tables";
1035 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1036 case E820_TYPE_UNUSABLE: return "Unusable memory";
1037 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1038 case E820_TYPE_PMEM: return "Persistent Memory";
1039 case E820_TYPE_RESERVED: return "Reserved";
1040 default: return "Unknown E820 type";
1041 }
1042}
1043
1044static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1045{
1046 switch (entry->type) {
1047 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1048 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1049 case E820_TYPE_ACPI: /* Fall-through: */
1050 case E820_TYPE_NVS: /* Fall-through: */
1051 case E820_TYPE_UNUSABLE: /* Fall-through: */
1052 case E820_TYPE_PRAM: /* Fall-through: */
1053 case E820_TYPE_PMEM: /* Fall-through: */
1054 case E820_TYPE_RESERVED: /* Fall-through: */
1055 default: return IORESOURCE_MEM;
1056 }
1057}
1058
1059static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1060{
1061 switch (entry->type) {
1062 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1063 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1064 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1065 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1066 case E820_TYPE_RESERVED: return IORES_DESC_RESERVED;
1067 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1068 case E820_TYPE_RAM: /* Fall-through: */
1069 case E820_TYPE_UNUSABLE: /* Fall-through: */
1070 default: return IORES_DESC_NONE;
1071 }
1072}
1073
1074static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1075{
1076 /* this is the legacy bios/dos rom-shadow + mmio region */
1077 if (res->start < (1ULL<<20))
1078 return true;
1079
1080 /*
1081 * Treat persistent memory like device memory, i.e. reserve it
1082 * for exclusive use of a driver
1083 */
1084 switch (type) {
1085 case E820_TYPE_RESERVED:
1086 case E820_TYPE_PRAM:
1087 case E820_TYPE_PMEM:
1088 return false;
1089 case E820_TYPE_RESERVED_KERN:
1090 case E820_TYPE_RAM:
1091 case E820_TYPE_ACPI:
1092 case E820_TYPE_NVS:
1093 case E820_TYPE_UNUSABLE:
1094 default:
1095 return true;
1096 }
1097}
1098
1099/*
1100 * Mark E820 reserved areas as busy for the resource manager:
1101 */
1102
1103static struct resource __initdata *e820_res;
1104
1105void __init e820__reserve_resources(void)
1106{
1107 int i;
1108 struct resource *res;
1109 u64 end;
1110
1111 res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1112 SMP_CACHE_BYTES);
1113 if (!res)
1114 panic("%s: Failed to allocate %zu bytes\n", __func__,
1115 sizeof(*res) * e820_table->nr_entries);
1116 e820_res = res;
1117
1118 for (i = 0; i < e820_table->nr_entries; i++) {
1119 struct e820_entry *entry = e820_table->entries + i;
1120
1121 end = entry->addr + entry->size - 1;
1122 if (end != (resource_size_t)end) {
1123 res++;
1124 continue;
1125 }
1126 res->start = entry->addr;
1127 res->end = end;
1128 res->name = e820_type_to_string(entry);
1129 res->flags = e820_type_to_iomem_type(entry);
1130 res->desc = e820_type_to_iores_desc(entry);
1131
1132 /*
1133 * Don't register the region that could be conflicted with
1134 * PCI device BAR resources and insert them later in
1135 * pcibios_resource_survey():
1136 */
1137 if (do_mark_busy(entry->type, res)) {
1138 res->flags |= IORESOURCE_BUSY;
1139 insert_resource(&iomem_resource, res);
1140 }
1141 res++;
1142 }
1143
1144 /* Expose the bootloader-provided memory layout to the sysfs. */
1145 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1146 struct e820_entry *entry = e820_table_firmware->entries + i;
1147
1148 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1149 }
1150}
1151
1152/*
1153 * How much should we pad the end of RAM, depending on where it is?
1154 */
1155static unsigned long __init ram_alignment(resource_size_t pos)
1156{
1157 unsigned long mb = pos >> 20;
1158
1159 /* To 64kB in the first megabyte */
1160 if (!mb)
1161 return 64*1024;
1162
1163 /* To 1MB in the first 16MB */
1164 if (mb < 16)
1165 return 1024*1024;
1166
1167 /* To 64MB for anything above that */
1168 return 64*1024*1024;
1169}
1170
1171#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1172
1173void __init e820__reserve_resources_late(void)
1174{
1175 int i;
1176 struct resource *res;
1177
1178 res = e820_res;
1179 for (i = 0; i < e820_table->nr_entries; i++) {
1180 if (!res->parent && res->end)
1181 insert_resource_expand_to_fit(&iomem_resource, res);
1182 res++;
1183 }
1184
1185 /*
1186 * Try to bump up RAM regions to reasonable boundaries, to
1187 * avoid stolen RAM:
1188 */
1189 for (i = 0; i < e820_table->nr_entries; i++) {
1190 struct e820_entry *entry = &e820_table->entries[i];
1191 u64 start, end;
1192
1193 if (entry->type != E820_TYPE_RAM)
1194 continue;
1195
1196 start = entry->addr + entry->size;
1197 end = round_up(start, ram_alignment(start)) - 1;
1198 if (end > MAX_RESOURCE_SIZE)
1199 end = MAX_RESOURCE_SIZE;
1200 if (start >= end)
1201 continue;
1202
1203 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1204 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1205 }
1206}
1207
1208/*
1209 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1210 */
1211char *__init e820__memory_setup_default(void)
1212{
1213 char *who = "BIOS-e820";
1214
1215 /*
1216 * Try to copy the BIOS-supplied E820-map.
1217 *
1218 * Otherwise fake a memory map; one section from 0k->640k,
1219 * the next section from 1mb->appropriate_mem_k
1220 */
1221 if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1222 u64 mem_size;
1223
1224 /* Compare results from other methods and take the one that gives more RAM: */
1225 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1226 mem_size = boot_params.screen_info.ext_mem_k;
1227 who = "BIOS-88";
1228 } else {
1229 mem_size = boot_params.alt_mem_k;
1230 who = "BIOS-e801";
1231 }
1232
1233 e820_table->nr_entries = 0;
1234 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1235 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1236 }
1237
1238 /* We just appended a lot of ranges, sanitize the table: */
1239 e820__update_table(e820_table);
1240
1241 return who;
1242}
1243
1244/*
1245 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1246 * E820 map - with an optional platform quirk available for virtual platforms
1247 * to override this method of boot environment processing:
1248 */
1249void __init e820__memory_setup(void)
1250{
1251 char *who;
1252
1253 /* This is a firmware interface ABI - make sure we don't break it: */
1254 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1255
1256 who = x86_init.resources.memory_setup();
1257
1258 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1259 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1260
1261 pr_info("BIOS-provided physical RAM map:\n");
1262 e820__print_table(who);
1263}
1264
1265void __init e820__memblock_setup(void)
1266{
1267 int i;
1268 u64 end;
1269
1270 /*
1271 * The bootstrap memblock region count maximum is 128 entries
1272 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1273 * than that - so allow memblock resizing.
1274 *
1275 * This is safe, because this call happens pretty late during x86 setup,
1276 * so we know about reserved memory regions already. (This is important
1277 * so that memblock resizing does no stomp over reserved areas.)
1278 */
1279 memblock_allow_resize();
1280
1281 for (i = 0; i < e820_table->nr_entries; i++) {
1282 struct e820_entry *entry = &e820_table->entries[i];
1283
1284 end = entry->addr + entry->size;
1285 if (end != (resource_size_t)end)
1286 continue;
1287
1288 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1289 continue;
1290
1291 memblock_add(entry->addr, entry->size);
1292 }
1293
1294 /* Throw away partial pages: */
1295 memblock_trim_memory(PAGE_SIZE);
1296
1297 memblock_dump_all();
1298}
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(e820_table_firmware, start, end, type);
97}
98EXPORT_SYMBOL_GPL(e820__mapped_raw_any);
99
100bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
101{
102 return _e820__mapped_any(e820_table, start, end, type);
103}
104EXPORT_SYMBOL_GPL(e820__mapped_any);
105
106/*
107 * This function checks if the entire <start,end> range is mapped with 'type'.
108 *
109 * Note: this function only works correctly once the E820 table is sorted and
110 * not-overlapping (at least for the range specified), which is the case normally.
111 */
112static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
113 enum e820_type type)
114{
115 int i;
116
117 for (i = 0; i < e820_table->nr_entries; i++) {
118 struct e820_entry *entry = &e820_table->entries[i];
119
120 if (type && entry->type != type)
121 continue;
122
123 /* Is the region (part) in overlap with the current region? */
124 if (entry->addr >= end || entry->addr + entry->size <= start)
125 continue;
126
127 /*
128 * If the region is at the beginning of <start,end> we move
129 * 'start' to the end of the region since it's ok until there
130 */
131 if (entry->addr <= start)
132 start = entry->addr + entry->size;
133
134 /*
135 * If 'start' is now at or beyond 'end', we're done, full
136 * coverage of the desired range exists:
137 */
138 if (start >= end)
139 return entry;
140 }
141
142 return NULL;
143}
144
145/*
146 * This function checks if the entire range <start,end> is mapped with type.
147 */
148bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
149{
150 return __e820__mapped_all(start, end, type);
151}
152
153/*
154 * This function returns the type associated with the range <start,end>.
155 */
156int e820__get_entry_type(u64 start, u64 end)
157{
158 struct e820_entry *entry = __e820__mapped_all(start, end, 0);
159
160 return entry ? entry->type : -EINVAL;
161}
162
163/*
164 * Add a memory region to the kernel E820 map.
165 */
166static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
167{
168 int x = table->nr_entries;
169
170 if (x >= ARRAY_SIZE(table->entries)) {
171 pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
172 start, start + size - 1);
173 return;
174 }
175
176 table->entries[x].addr = start;
177 table->entries[x].size = size;
178 table->entries[x].type = type;
179 table->nr_entries++;
180}
181
182void __init e820__range_add(u64 start, u64 size, enum e820_type type)
183{
184 __e820__range_add(e820_table, start, size, type);
185}
186
187static void __init e820_print_type(enum e820_type type)
188{
189 switch (type) {
190 case E820_TYPE_RAM: /* Fall through: */
191 case E820_TYPE_RESERVED_KERN: pr_cont("usable"); break;
192 case E820_TYPE_RESERVED: pr_cont("reserved"); break;
193 case E820_TYPE_SOFT_RESERVED: pr_cont("soft reserved"); break;
194 case E820_TYPE_ACPI: pr_cont("ACPI data"); break;
195 case E820_TYPE_NVS: pr_cont("ACPI NVS"); break;
196 case E820_TYPE_UNUSABLE: pr_cont("unusable"); break;
197 case E820_TYPE_PMEM: /* Fall through: */
198 case E820_TYPE_PRAM: pr_cont("persistent (type %u)", type); break;
199 default: pr_cont("type %u", type); break;
200 }
201}
202
203void __init e820__print_table(char *who)
204{
205 int i;
206
207 for (i = 0; i < e820_table->nr_entries; i++) {
208 pr_info("%s: [mem %#018Lx-%#018Lx] ",
209 who,
210 e820_table->entries[i].addr,
211 e820_table->entries[i].addr + e820_table->entries[i].size - 1);
212
213 e820_print_type(e820_table->entries[i].type);
214 pr_cont("\n");
215 }
216}
217
218/*
219 * Sanitize an E820 map.
220 *
221 * Some E820 layouts include overlapping entries. The following
222 * replaces the original E820 map with a new one, removing overlaps,
223 * and resolving conflicting memory types in favor of highest
224 * numbered type.
225 *
226 * The input parameter 'entries' points to an array of 'struct
227 * e820_entry' which on entry has elements in the range [0, *nr_entries)
228 * valid, and which has space for up to max_nr_entries entries.
229 * On return, the resulting sanitized E820 map entries will be in
230 * overwritten in the same location, starting at 'entries'.
231 *
232 * The integer pointed to by nr_entries must be valid on entry (the
233 * current number of valid entries located at 'entries'). If the
234 * sanitizing succeeds the *nr_entries will be updated with the new
235 * number of valid entries (something no more than max_nr_entries).
236 *
237 * The return value from e820__update_table() is zero if it
238 * successfully 'sanitized' the map entries passed in, and is -1
239 * if it did nothing, which can happen if either of (1) it was
240 * only passed one map entry, or (2) any of the input map entries
241 * were invalid (start + size < start, meaning that the size was
242 * so big the described memory range wrapped around through zero.)
243 *
244 * Visually we're performing the following
245 * (1,2,3,4 = memory types)...
246 *
247 * Sample memory map (w/overlaps):
248 * ____22__________________
249 * ______________________4_
250 * ____1111________________
251 * _44_____________________
252 * 11111111________________
253 * ____________________33__
254 * ___________44___________
255 * __________33333_________
256 * ______________22________
257 * ___________________2222_
258 * _________111111111______
259 * _____________________11_
260 * _________________4______
261 *
262 * Sanitized equivalent (no overlap):
263 * 1_______________________
264 * _44_____________________
265 * ___1____________________
266 * ____22__________________
267 * ______11________________
268 * _________1______________
269 * __________3_____________
270 * ___________44___________
271 * _____________33_________
272 * _______________2________
273 * ________________1_______
274 * _________________4______
275 * ___________________2____
276 * ____________________33__
277 * ______________________4_
278 */
279struct change_member {
280 /* Pointer to the original entry: */
281 struct e820_entry *entry;
282 /* Address for this change point: */
283 unsigned long long addr;
284};
285
286static struct change_member change_point_list[2*E820_MAX_ENTRIES] __initdata;
287static struct change_member *change_point[2*E820_MAX_ENTRIES] __initdata;
288static struct e820_entry *overlap_list[E820_MAX_ENTRIES] __initdata;
289static struct e820_entry new_entries[E820_MAX_ENTRIES] __initdata;
290
291static int __init cpcompare(const void *a, const void *b)
292{
293 struct change_member * const *app = a, * const *bpp = b;
294 const struct change_member *ap = *app, *bp = *bpp;
295
296 /*
297 * Inputs are pointers to two elements of change_point[]. If their
298 * addresses are not equal, their difference dominates. If the addresses
299 * are equal, then consider one that represents the end of its region
300 * to be greater than one that does not.
301 */
302 if (ap->addr != bp->addr)
303 return ap->addr > bp->addr ? 1 : -1;
304
305 return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
306}
307
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(change_point, chg_nr, sizeof(*change_point), 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(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(old_type);
477 pr_cont(" ==> ");
478 e820_print_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, new_type);
501 __e820__range_add(table, end, entry_end - end, 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, final_start, final_end - final_start, 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(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(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(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(end, entry_end - end, 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(e820_table))
605 return;
606
607 pr_info("modified physical RAM map:\n");
608 e820__print_table("modified");
609}
610
611static void __init e820__update_table_kexec(void)
612{
613 e820__update_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, &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, data_len);
731 entries = sdata->len / sizeof(*extmap);
732 extmap = (struct boot_e820_entry *)(sdata->data);
733
734 __append_e820_table(extmap, entries);
735 e820__update_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(sdata, data_len);
741 pr_info("extended physical RAM map:\n");
742 e820__print_table("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(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), 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(entry->addr, 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(addr, size, E820_TYPE_RAM, 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, E820_TYPE_RAM);
868}
869
870unsigned long __init e820__end_of_low_ram_pfn(void)
871{
872 return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
873}
874
875static void __init early_panic(char *msg)
876{
877 early_printk(msg);
878 panic(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(p, &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(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 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(p, &p);
934 if (p == oldp)
935 return -EINVAL;
936
937 userdef = 1;
938 if (*p == '@') {
939 start_at = memparse(p+1, &p);
940 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
941 } else if (*p == '#') {
942 start_at = memparse(p+1, &p);
943 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
944 } else if (*p == '$') {
945 start_at = memparse(p+1, &p);
946 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
947 } else if (*p == '!') {
948 start_at = memparse(p+1, &p);
949 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
950 } else if (*p == '%') {
951 enum e820_type from = 0, to = 0;
952
953 start_at = memparse(p + 1, &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_at, mem_size, from, to);
962 else if (to)
963 e820__range_add(start_at, mem_size, to);
964 else if (from)
965 e820__range_remove(start_at, mem_size, from, 1);
966 else
967 e820__range_remove(start_at, mem_size, 0, 0);
968 } else {
969 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 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(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(pa_data, 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(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1018
1019 /*
1020 * SETUP_EFI, SETUP_IMA and SETUP_RNG_SEED are supplied by
1021 * kexec and do not need to be reserved.
1022 */
1023 if (data->type != SETUP_EFI &&
1024 data->type != SETUP_IMA &&
1025 data->type != SETUP_RNG_SEED)
1026 e820__range_update_kexec(pa_data,
1027 sizeof(*data) + data->len,
1028 E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1029
1030 if (data->type == SETUP_INDIRECT) {
1031 len += data->len;
1032 early_memunmap(data, sizeof(*data));
1033 data = early_memremap(pa_data, len);
1034 if (!data) {
1035 pr_warn("e820: failed to memremap indirect setup_data\n");
1036 return;
1037 }
1038
1039 indirect = (struct setup_indirect *)data->data;
1040
1041 if (indirect->type != SETUP_INDIRECT) {
1042 e820__range_update(indirect->addr, indirect->len,
1043 E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1044 e820__range_update_kexec(indirect->addr, indirect->len,
1045 E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1046 }
1047 }
1048
1049 pa_data = pa_next;
1050 early_memunmap(data, len);
1051 }
1052
1053 e820__update_table(e820_table);
1054 e820__update_table(e820_table_kexec);
1055
1056 pr_info("extended physical RAM map:\n");
1057 e820__print_table("reserve setup_data");
1058}
1059
1060/*
1061 * Called after parse_early_param(), after early parameters (such as mem=)
1062 * have been processed, in which case we already have an E820 table filled in
1063 * via the parameter callback function(s), but it's not sorted and printed yet:
1064 */
1065void __init e820__finish_early_params(void)
1066{
1067 if (userdef) {
1068 if (e820__update_table(e820_table) < 0)
1069 early_panic("Invalid user supplied memory map");
1070
1071 pr_info("user-defined physical RAM map:\n");
1072 e820__print_table("user");
1073 }
1074}
1075
1076static const char *__init e820_type_to_string(struct e820_entry *entry)
1077{
1078 switch (entry->type) {
1079 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1080 case E820_TYPE_RAM: return "System RAM";
1081 case E820_TYPE_ACPI: return "ACPI Tables";
1082 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1083 case E820_TYPE_UNUSABLE: return "Unusable memory";
1084 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1085 case E820_TYPE_PMEM: return "Persistent Memory";
1086 case E820_TYPE_RESERVED: return "Reserved";
1087 case E820_TYPE_SOFT_RESERVED: return "Soft Reserved";
1088 default: return "Unknown E820 type";
1089 }
1090}
1091
1092static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1093{
1094 switch (entry->type) {
1095 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1096 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1097 case E820_TYPE_ACPI: /* Fall-through: */
1098 case E820_TYPE_NVS: /* Fall-through: */
1099 case E820_TYPE_UNUSABLE: /* Fall-through: */
1100 case E820_TYPE_PRAM: /* Fall-through: */
1101 case E820_TYPE_PMEM: /* Fall-through: */
1102 case E820_TYPE_RESERVED: /* Fall-through: */
1103 case E820_TYPE_SOFT_RESERVED: /* Fall-through: */
1104 default: return IORESOURCE_MEM;
1105 }
1106}
1107
1108static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1109{
1110 switch (entry->type) {
1111 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1112 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1113 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1114 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1115 case E820_TYPE_RESERVED: return IORES_DESC_RESERVED;
1116 case E820_TYPE_SOFT_RESERVED: return IORES_DESC_SOFT_RESERVED;
1117 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1118 case E820_TYPE_RAM: /* Fall-through: */
1119 case E820_TYPE_UNUSABLE: /* Fall-through: */
1120 default: return IORES_DESC_NONE;
1121 }
1122}
1123
1124static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1125{
1126 /* this is the legacy bios/dos rom-shadow + mmio region */
1127 if (res->start < (1ULL<<20))
1128 return true;
1129
1130 /*
1131 * Treat persistent memory and other special memory ranges like
1132 * device memory, i.e. reserve it for exclusive use of a driver
1133 */
1134 switch (type) {
1135 case E820_TYPE_RESERVED:
1136 case E820_TYPE_SOFT_RESERVED:
1137 case E820_TYPE_PRAM:
1138 case E820_TYPE_PMEM:
1139 return false;
1140 case E820_TYPE_RESERVED_KERN:
1141 case E820_TYPE_RAM:
1142 case E820_TYPE_ACPI:
1143 case E820_TYPE_NVS:
1144 case E820_TYPE_UNUSABLE:
1145 default:
1146 return true;
1147 }
1148}
1149
1150/*
1151 * Mark E820 reserved areas as busy for the resource manager:
1152 */
1153
1154static struct resource __initdata *e820_res;
1155
1156void __init e820__reserve_resources(void)
1157{
1158 int i;
1159 struct resource *res;
1160 u64 end;
1161
1162 res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1163 SMP_CACHE_BYTES);
1164 if (!res)
1165 panic("%s: Failed to allocate %zu bytes\n", __func__,
1166 sizeof(*res) * e820_table->nr_entries);
1167 e820_res = res;
1168
1169 for (i = 0; i < e820_table->nr_entries; i++) {
1170 struct e820_entry *entry = e820_table->entries + i;
1171
1172 end = entry->addr + entry->size - 1;
1173 if (end != (resource_size_t)end) {
1174 res++;
1175 continue;
1176 }
1177 res->start = entry->addr;
1178 res->end = end;
1179 res->name = e820_type_to_string(entry);
1180 res->flags = e820_type_to_iomem_type(entry);
1181 res->desc = e820_type_to_iores_desc(entry);
1182
1183 /*
1184 * Don't register the region that could be conflicted with
1185 * PCI device BAR resources and insert them later in
1186 * pcibios_resource_survey():
1187 */
1188 if (do_mark_busy(entry->type, res)) {
1189 res->flags |= IORESOURCE_BUSY;
1190 insert_resource(&iomem_resource, res);
1191 }
1192 res++;
1193 }
1194
1195 /* Expose the bootloader-provided memory layout to the sysfs. */
1196 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1197 struct e820_entry *entry = e820_table_firmware->entries + i;
1198
1199 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1200 }
1201}
1202
1203/*
1204 * How much should we pad the end of RAM, depending on where it is?
1205 */
1206static unsigned long __init ram_alignment(resource_size_t pos)
1207{
1208 unsigned long mb = pos >> 20;
1209
1210 /* To 64kB in the first megabyte */
1211 if (!mb)
1212 return 64*1024;
1213
1214 /* To 1MB in the first 16MB */
1215 if (mb < 16)
1216 return 1024*1024;
1217
1218 /* To 64MB for anything above that */
1219 return 64*1024*1024;
1220}
1221
1222#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1223
1224void __init e820__reserve_resources_late(void)
1225{
1226 int i;
1227 struct resource *res;
1228
1229 res = e820_res;
1230 for (i = 0; i < e820_table->nr_entries; i++) {
1231 if (!res->parent && res->end)
1232 insert_resource_expand_to_fit(&iomem_resource, res);
1233 res++;
1234 }
1235
1236 /*
1237 * Try to bump up RAM regions to reasonable boundaries, to
1238 * avoid stolen RAM:
1239 */
1240 for (i = 0; i < e820_table->nr_entries; i++) {
1241 struct e820_entry *entry = &e820_table->entries[i];
1242 u64 start, end;
1243
1244 if (entry->type != E820_TYPE_RAM)
1245 continue;
1246
1247 start = entry->addr + entry->size;
1248 end = round_up(start, ram_alignment(start)) - 1;
1249 if (end > MAX_RESOURCE_SIZE)
1250 end = MAX_RESOURCE_SIZE;
1251 if (start >= end)
1252 continue;
1253
1254 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1255 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1256 }
1257}
1258
1259/*
1260 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1261 */
1262char *__init e820__memory_setup_default(void)
1263{
1264 char *who = "BIOS-e820";
1265
1266 /*
1267 * Try to copy the BIOS-supplied E820-map.
1268 *
1269 * Otherwise fake a memory map; one section from 0k->640k,
1270 * the next section from 1mb->appropriate_mem_k
1271 */
1272 if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1273 u64 mem_size;
1274
1275 /* Compare results from other methods and take the one that gives more RAM: */
1276 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1277 mem_size = boot_params.screen_info.ext_mem_k;
1278 who = "BIOS-88";
1279 } else {
1280 mem_size = boot_params.alt_mem_k;
1281 who = "BIOS-e801";
1282 }
1283
1284 e820_table->nr_entries = 0;
1285 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1286 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1287 }
1288
1289 /* We just appended a lot of ranges, sanitize the table: */
1290 e820__update_table(e820_table);
1291
1292 return who;
1293}
1294
1295/*
1296 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1297 * E820 map - with an optional platform quirk available for virtual platforms
1298 * to override this method of boot environment processing:
1299 */
1300void __init e820__memory_setup(void)
1301{
1302 char *who;
1303
1304 /* This is a firmware interface ABI - make sure we don't break it: */
1305 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1306
1307 who = x86_init.resources.memory_setup();
1308
1309 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1310 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1311
1312 pr_info("BIOS-provided physical RAM map:\n");
1313 e820__print_table(who);
1314}
1315
1316void __init e820__memblock_setup(void)
1317{
1318 int i;
1319 u64 end;
1320
1321 /*
1322 * The bootstrap memblock region count maximum is 128 entries
1323 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1324 * than that - so allow memblock resizing.
1325 *
1326 * This is safe, because this call happens pretty late during x86 setup,
1327 * so we know about reserved memory regions already. (This is important
1328 * so that memblock resizing does no stomp over reserved areas.)
1329 */
1330 memblock_allow_resize();
1331
1332 for (i = 0; i < e820_table->nr_entries; i++) {
1333 struct e820_entry *entry = &e820_table->entries[i];
1334
1335 end = entry->addr + entry->size;
1336 if (end != (resource_size_t)end)
1337 continue;
1338
1339 if (entry->type == E820_TYPE_SOFT_RESERVED)
1340 memblock_reserve(entry->addr, entry->size);
1341
1342 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1343 continue;
1344
1345 memblock_add(entry->addr, entry->size);
1346 }
1347
1348 /* Throw away partial pages: */
1349 memblock_trim_memory(PAGE_SIZE);
1350
1351 memblock_dump_all();
1352}