Loading...
1/*
2 * Low level x86 E820 memory map handling functions.
3 *
4 * The firmware and bootloader passes us the "E820 table", which is the primary
5 * physical memory layout description available about x86 systems.
6 *
7 * The kernel takes the E820 memory layout and optionally modifies it with
8 * quirks and other tweaks, and feeds that into the generic Linux memory
9 * allocation code routines via a platform independent interface (memblock, etc.).
10 */
11#include <linux/crash_dump.h>
12#include <linux/bootmem.h>
13#include <linux/suspend.h>
14#include <linux/acpi.h>
15#include <linux/firmware-map.h>
16#include <linux/memblock.h>
17#include <linux/sort.h>
18
19#include <asm/e820/api.h>
20#include <asm/setup.h>
21
22/*
23 * We organize the E820 table into three main data structures:
24 *
25 * - 'e820_table_firmware': the original firmware version passed to us by the
26 * bootloader - not modified by the kernel. It is composed of two parts:
27 * the first 128 E820 memory entries in boot_params.e820_table and the remaining
28 * (if any) entries of the SETUP_E820_EXT nodes. We use this to:
29 *
30 * - inform the user about the firmware's notion of memory layout
31 * via /sys/firmware/memmap
32 *
33 * - the hibernation code uses it to generate a kernel-independent MD5
34 * fingerprint of the physical memory layout of a system.
35 *
36 * - 'e820_table_kexec': a slightly modified (by the kernel) firmware version
37 * passed to us by the bootloader - the major difference between
38 * e820_table_firmware[] and this one is that, the latter marks the setup_data
39 * list created by the EFI boot stub as reserved, so that kexec can reuse the
40 * setup_data information in the second kernel. Besides, e820_table_kexec[]
41 * might also be modified by the kexec itself to fake a mptable.
42 * We use this to:
43 *
44 * - kexec, which is a bootloader in disguise, uses the original E820
45 * layout to pass to the kexec-ed kernel. This way the original kernel
46 * can have a restricted E820 map while the kexec()-ed kexec-kernel
47 * can have access to full memory - etc.
48 *
49 * - 'e820_table': this is the main E820 table that is massaged by the
50 * low level x86 platform code, or modified by boot parameters, before
51 * passed on to higher level MM layers.
52 *
53 * Once the E820 map has been converted to the standard Linux memory layout
54 * information its role stops - modifying it has no effect and does not get
55 * re-propagated. So itsmain role is a temporary bootstrap storage of firmware
56 * specific memory layout data during early bootup.
57 */
58static struct e820_table e820_table_init __initdata;
59static struct e820_table e820_table_kexec_init __initdata;
60static struct e820_table e820_table_firmware_init __initdata;
61
62struct e820_table *e820_table __refdata = &e820_table_init;
63struct e820_table *e820_table_kexec __refdata = &e820_table_kexec_init;
64struct e820_table *e820_table_firmware __refdata = &e820_table_firmware_init;
65
66/* For PCI or other memory-mapped resources */
67unsigned long pci_mem_start = 0xaeedbabe;
68#ifdef CONFIG_PCI
69EXPORT_SYMBOL(pci_mem_start);
70#endif
71
72/*
73 * This function checks if any part of the range <start,end> is mapped
74 * with type.
75 */
76bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
77{
78 int i;
79
80 for (i = 0; i < e820_table->nr_entries; i++) {
81 struct e820_entry *entry = &e820_table->entries[i];
82
83 if (type && entry->type != type)
84 continue;
85 if (entry->addr >= end || entry->addr + entry->size <= start)
86 continue;
87 return 1;
88 }
89 return 0;
90}
91EXPORT_SYMBOL_GPL(e820__mapped_any);
92
93/*
94 * This function checks if the entire <start,end> range is mapped with 'type'.
95 *
96 * Note: this function only works correctly once the E820 table is sorted and
97 * not-overlapping (at least for the range specified), which is the case normally.
98 */
99static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
100 enum e820_type type)
101{
102 int i;
103
104 for (i = 0; i < e820_table->nr_entries; i++) {
105 struct e820_entry *entry = &e820_table->entries[i];
106
107 if (type && entry->type != type)
108 continue;
109
110 /* Is the region (part) in overlap with the current region? */
111 if (entry->addr >= end || entry->addr + entry->size <= start)
112 continue;
113
114 /*
115 * If the region is at the beginning of <start,end> we move
116 * 'start' to the end of the region since it's ok until there
117 */
118 if (entry->addr <= start)
119 start = entry->addr + entry->size;
120
121 /*
122 * If 'start' is now at or beyond 'end', we're done, full
123 * coverage of the desired range exists:
124 */
125 if (start >= end)
126 return entry;
127 }
128
129 return NULL;
130}
131
132/*
133 * This function checks if the entire range <start,end> is mapped with type.
134 */
135bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
136{
137 return __e820__mapped_all(start, end, type);
138}
139
140/*
141 * This function returns the type associated with the range <start,end>.
142 */
143int e820__get_entry_type(u64 start, u64 end)
144{
145 struct e820_entry *entry = __e820__mapped_all(start, end, 0);
146
147 return entry ? entry->type : -EINVAL;
148}
149
150/*
151 * Add a memory region to the kernel E820 map.
152 */
153static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
154{
155 int x = table->nr_entries;
156
157 if (x >= ARRAY_SIZE(table->entries)) {
158 pr_err("e820: too many entries; ignoring [mem %#010llx-%#010llx]\n", start, start + size - 1);
159 return;
160 }
161
162 table->entries[x].addr = start;
163 table->entries[x].size = size;
164 table->entries[x].type = type;
165 table->nr_entries++;
166}
167
168void __init e820__range_add(u64 start, u64 size, enum e820_type type)
169{
170 __e820__range_add(e820_table, start, size, type);
171}
172
173static void __init e820_print_type(enum e820_type type)
174{
175 switch (type) {
176 case E820_TYPE_RAM: /* Fall through: */
177 case E820_TYPE_RESERVED_KERN: pr_cont("usable"); break;
178 case E820_TYPE_RESERVED: pr_cont("reserved"); break;
179 case E820_TYPE_ACPI: pr_cont("ACPI data"); break;
180 case E820_TYPE_NVS: pr_cont("ACPI NVS"); break;
181 case E820_TYPE_UNUSABLE: pr_cont("unusable"); break;
182 case E820_TYPE_PMEM: /* Fall through: */
183 case E820_TYPE_PRAM: pr_cont("persistent (type %u)", type); break;
184 default: pr_cont("type %u", type); break;
185 }
186}
187
188void __init e820__print_table(char *who)
189{
190 int i;
191
192 for (i = 0; i < e820_table->nr_entries; i++) {
193 pr_info("%s: [mem %#018Lx-%#018Lx] ", who,
194 e820_table->entries[i].addr,
195 e820_table->entries[i].addr + e820_table->entries[i].size - 1);
196
197 e820_print_type(e820_table->entries[i].type);
198 pr_cont("\n");
199 }
200}
201
202/*
203 * Sanitize an E820 map.
204 *
205 * Some E820 layouts include overlapping entries. The following
206 * replaces the original E820 map with a new one, removing overlaps,
207 * and resolving conflicting memory types in favor of highest
208 * numbered type.
209 *
210 * The input parameter 'entries' points to an array of 'struct
211 * e820_entry' which on entry has elements in the range [0, *nr_entries)
212 * valid, and which has space for up to max_nr_entries entries.
213 * On return, the resulting sanitized E820 map entries will be in
214 * overwritten in the same location, starting at 'entries'.
215 *
216 * The integer pointed to by nr_entries must be valid on entry (the
217 * current number of valid entries located at 'entries'). If the
218 * sanitizing succeeds the *nr_entries will be updated with the new
219 * number of valid entries (something no more than max_nr_entries).
220 *
221 * The return value from e820__update_table() is zero if it
222 * successfully 'sanitized' the map entries passed in, and is -1
223 * if it did nothing, which can happen if either of (1) it was
224 * only passed one map entry, or (2) any of the input map entries
225 * were invalid (start + size < start, meaning that the size was
226 * so big the described memory range wrapped around through zero.)
227 *
228 * Visually we're performing the following
229 * (1,2,3,4 = memory types)...
230 *
231 * Sample memory map (w/overlaps):
232 * ____22__________________
233 * ______________________4_
234 * ____1111________________
235 * _44_____________________
236 * 11111111________________
237 * ____________________33__
238 * ___________44___________
239 * __________33333_________
240 * ______________22________
241 * ___________________2222_
242 * _________111111111______
243 * _____________________11_
244 * _________________4______
245 *
246 * Sanitized equivalent (no overlap):
247 * 1_______________________
248 * _44_____________________
249 * ___1____________________
250 * ____22__________________
251 * ______11________________
252 * _________1______________
253 * __________3_____________
254 * ___________44___________
255 * _____________33_________
256 * _______________2________
257 * ________________1_______
258 * _________________4______
259 * ___________________2____
260 * ____________________33__
261 * ______________________4_
262 */
263struct change_member {
264 /* Pointer to the original entry: */
265 struct e820_entry *entry;
266 /* Address for this change point: */
267 unsigned long long addr;
268};
269
270static struct change_member change_point_list[2*E820_MAX_ENTRIES] __initdata;
271static struct change_member *change_point[2*E820_MAX_ENTRIES] __initdata;
272static struct e820_entry *overlap_list[E820_MAX_ENTRIES] __initdata;
273static struct e820_entry new_entries[E820_MAX_ENTRIES] __initdata;
274
275static int __init cpcompare(const void *a, const void *b)
276{
277 struct change_member * const *app = a, * const *bpp = b;
278 const struct change_member *ap = *app, *bp = *bpp;
279
280 /*
281 * Inputs are pointers to two elements of change_point[]. If their
282 * addresses are not equal, their difference dominates. If the addresses
283 * are equal, then consider one that represents the end of its region
284 * to be greater than one that does not.
285 */
286 if (ap->addr != bp->addr)
287 return ap->addr > bp->addr ? 1 : -1;
288
289 return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
290}
291
292int __init e820__update_table(struct e820_table *table)
293{
294 struct e820_entry *entries = table->entries;
295 u32 max_nr_entries = ARRAY_SIZE(table->entries);
296 enum e820_type current_type, last_type;
297 unsigned long long last_addr;
298 u32 new_nr_entries, overlap_entries;
299 u32 i, chg_idx, chg_nr;
300
301 /* If there's only one memory region, don't bother: */
302 if (table->nr_entries < 2)
303 return -1;
304
305 BUG_ON(table->nr_entries > max_nr_entries);
306
307 /* Bail out if we find any unreasonable addresses in the map: */
308 for (i = 0; i < table->nr_entries; i++) {
309 if (entries[i].addr + entries[i].size < entries[i].addr)
310 return -1;
311 }
312
313 /* Create pointers for initial change-point information (for sorting): */
314 for (i = 0; i < 2 * table->nr_entries; i++)
315 change_point[i] = &change_point_list[i];
316
317 /*
318 * Record all known change-points (starting and ending addresses),
319 * omitting empty memory regions:
320 */
321 chg_idx = 0;
322 for (i = 0; i < table->nr_entries; i++) {
323 if (entries[i].size != 0) {
324 change_point[chg_idx]->addr = entries[i].addr;
325 change_point[chg_idx++]->entry = &entries[i];
326 change_point[chg_idx]->addr = entries[i].addr + entries[i].size;
327 change_point[chg_idx++]->entry = &entries[i];
328 }
329 }
330 chg_nr = chg_idx;
331
332 /* Sort change-point list by memory addresses (low -> high): */
333 sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
334
335 /* Create a new memory map, removing overlaps: */
336 overlap_entries = 0; /* Number of entries in the overlap table */
337 new_nr_entries = 0; /* Index for creating new map entries */
338 last_type = 0; /* Start with undefined memory type */
339 last_addr = 0; /* Start with 0 as last starting address */
340
341 /* Loop through change-points, determining effect on the new map: */
342 for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
343 /* Keep track of all overlapping entries */
344 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
345 /* Add map entry to overlap list (> 1 entry implies an overlap) */
346 overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
347 } else {
348 /* Remove entry from list (order independent, so swap with last): */
349 for (i = 0; i < overlap_entries; i++) {
350 if (overlap_list[i] == change_point[chg_idx]->entry)
351 overlap_list[i] = overlap_list[overlap_entries-1];
352 }
353 overlap_entries--;
354 }
355 /*
356 * If there are overlapping entries, decide which
357 * "type" to use (larger value takes precedence --
358 * 1=usable, 2,3,4,4+=unusable)
359 */
360 current_type = 0;
361 for (i = 0; i < overlap_entries; i++) {
362 if (overlap_list[i]->type > current_type)
363 current_type = overlap_list[i]->type;
364 }
365
366 /* Continue building up new map based on this information: */
367 if (current_type != last_type || current_type == E820_TYPE_PRAM) {
368 if (last_type != 0) {
369 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
370 /* Move forward only if the new size was non-zero: */
371 if (new_entries[new_nr_entries].size != 0)
372 /* No more space left for new entries? */
373 if (++new_nr_entries >= max_nr_entries)
374 break;
375 }
376 if (current_type != 0) {
377 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
378 new_entries[new_nr_entries].type = current_type;
379 last_addr = change_point[chg_idx]->addr;
380 }
381 last_type = current_type;
382 }
383 }
384
385 /* Copy the new entries into the original location: */
386 memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
387 table->nr_entries = new_nr_entries;
388
389 return 0;
390}
391
392static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
393{
394 struct boot_e820_entry *entry = entries;
395
396 while (nr_entries) {
397 u64 start = entry->addr;
398 u64 size = entry->size;
399 u64 end = start + size - 1;
400 u32 type = entry->type;
401
402 /* Ignore the entry on 64-bit overflow: */
403 if (start > end && likely(size))
404 return -1;
405
406 e820__range_add(start, size, type);
407
408 entry++;
409 nr_entries--;
410 }
411 return 0;
412}
413
414/*
415 * Copy the BIOS E820 map into a safe place.
416 *
417 * Sanity-check it while we're at it..
418 *
419 * If we're lucky and live on a modern system, the setup code
420 * will have given us a memory map that we can use to properly
421 * set up memory. If we aren't, we'll fake a memory map.
422 */
423static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
424{
425 /* Only one memory region (or negative)? Ignore it */
426 if (nr_entries < 2)
427 return -1;
428
429 return __append_e820_table(entries, nr_entries);
430}
431
432static u64 __init
433__e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
434{
435 u64 end;
436 unsigned int i;
437 u64 real_updated_size = 0;
438
439 BUG_ON(old_type == new_type);
440
441 if (size > (ULLONG_MAX - start))
442 size = ULLONG_MAX - start;
443
444 end = start + size;
445 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
446 e820_print_type(old_type);
447 pr_cont(" ==> ");
448 e820_print_type(new_type);
449 pr_cont("\n");
450
451 for (i = 0; i < table->nr_entries; i++) {
452 struct e820_entry *entry = &table->entries[i];
453 u64 final_start, final_end;
454 u64 entry_end;
455
456 if (entry->type != old_type)
457 continue;
458
459 entry_end = entry->addr + entry->size;
460
461 /* Completely covered by new range? */
462 if (entry->addr >= start && entry_end <= end) {
463 entry->type = new_type;
464 real_updated_size += entry->size;
465 continue;
466 }
467
468 /* New range is completely covered? */
469 if (entry->addr < start && entry_end > end) {
470 __e820__range_add(table, start, size, new_type);
471 __e820__range_add(table, end, entry_end - end, entry->type);
472 entry->size = start - entry->addr;
473 real_updated_size += size;
474 continue;
475 }
476
477 /* Partially covered: */
478 final_start = max(start, entry->addr);
479 final_end = min(end, entry_end);
480 if (final_start >= final_end)
481 continue;
482
483 __e820__range_add(table, final_start, final_end - final_start, new_type);
484
485 real_updated_size += final_end - final_start;
486
487 /*
488 * Left range could be head or tail, so need to update
489 * its size first:
490 */
491 entry->size -= final_end - final_start;
492 if (entry->addr < final_start)
493 continue;
494
495 entry->addr = final_end;
496 }
497 return real_updated_size;
498}
499
500u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
501{
502 return __e820__range_update(e820_table, start, size, old_type, new_type);
503}
504
505static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
506{
507 return __e820__range_update(e820_table_kexec, start, size, old_type, new_type);
508}
509
510/* Remove a range of memory from the E820 table: */
511u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
512{
513 int i;
514 u64 end;
515 u64 real_removed_size = 0;
516
517 if (size > (ULLONG_MAX - start))
518 size = ULLONG_MAX - start;
519
520 end = start + size;
521 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
522 if (check_type)
523 e820_print_type(old_type);
524 pr_cont("\n");
525
526 for (i = 0; i < e820_table->nr_entries; i++) {
527 struct e820_entry *entry = &e820_table->entries[i];
528 u64 final_start, final_end;
529 u64 entry_end;
530
531 if (check_type && entry->type != old_type)
532 continue;
533
534 entry_end = entry->addr + entry->size;
535
536 /* Completely covered? */
537 if (entry->addr >= start && entry_end <= end) {
538 real_removed_size += entry->size;
539 memset(entry, 0, sizeof(*entry));
540 continue;
541 }
542
543 /* Is the new range completely covered? */
544 if (entry->addr < start && entry_end > end) {
545 e820__range_add(end, entry_end - end, entry->type);
546 entry->size = start - entry->addr;
547 real_removed_size += size;
548 continue;
549 }
550
551 /* Partially covered: */
552 final_start = max(start, entry->addr);
553 final_end = min(end, entry_end);
554 if (final_start >= final_end)
555 continue;
556
557 real_removed_size += final_end - final_start;
558
559 /*
560 * Left range could be head or tail, so need to update
561 * the size first:
562 */
563 entry->size -= final_end - final_start;
564 if (entry->addr < final_start)
565 continue;
566
567 entry->addr = final_end;
568 }
569 return real_removed_size;
570}
571
572void __init e820__update_table_print(void)
573{
574 if (e820__update_table(e820_table))
575 return;
576
577 pr_info("e820: modified physical RAM map:\n");
578 e820__print_table("modified");
579}
580
581static void __init e820__update_table_kexec(void)
582{
583 e820__update_table(e820_table_kexec);
584}
585
586#define MAX_GAP_END 0x100000000ull
587
588/*
589 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
590 */
591static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
592{
593 unsigned long long last = MAX_GAP_END;
594 int i = e820_table->nr_entries;
595 int found = 0;
596
597 while (--i >= 0) {
598 unsigned long long start = e820_table->entries[i].addr;
599 unsigned long long end = start + e820_table->entries[i].size;
600
601 /*
602 * Since "last" is at most 4GB, we know we'll
603 * fit in 32 bits if this condition is true:
604 */
605 if (last > end) {
606 unsigned long gap = last - end;
607
608 if (gap >= *gapsize) {
609 *gapsize = gap;
610 *gapstart = end;
611 found = 1;
612 }
613 }
614 if (start < last)
615 last = start;
616 }
617 return found;
618}
619
620/*
621 * Search for the biggest gap in the low 32 bits of the E820
622 * memory space. We pass this space to the PCI subsystem, so
623 * that it can assign MMIO resources for hotplug or
624 * unconfigured devices in.
625 *
626 * Hopefully the BIOS let enough space left.
627 */
628__init void e820__setup_pci_gap(void)
629{
630 unsigned long gapstart, gapsize;
631 int found;
632
633 gapsize = 0x400000;
634 found = e820_search_gap(&gapstart, &gapsize);
635
636 if (!found) {
637#ifdef CONFIG_X86_64
638 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
639 pr_err(
640 "e820: Cannot find an available gap in the 32-bit address range\n"
641 "e820: PCI devices with unassigned 32-bit BARs may not work!\n");
642#else
643 gapstart = 0x10000000;
644#endif
645 }
646
647 /*
648 * e820__reserve_resources_late() protects stolen RAM already:
649 */
650 pci_mem_start = gapstart;
651
652 pr_info("e820: [mem %#010lx-%#010lx] available for PCI devices\n", gapstart, gapstart + gapsize - 1);
653}
654
655/*
656 * Called late during init, in free_initmem().
657 *
658 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
659 *
660 * Copy them to a (usually much smaller) dynamically allocated area that is
661 * sized precisely after the number of e820 entries.
662 *
663 * This is done after we've performed all the fixes and tweaks to the tables.
664 * All functions which modify them are __init functions, which won't exist
665 * after free_initmem().
666 */
667__init void e820__reallocate_tables(void)
668{
669 struct e820_table *n;
670 int size;
671
672 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
673 n = kmalloc(size, GFP_KERNEL);
674 BUG_ON(!n);
675 memcpy(n, e820_table, size);
676 e820_table = n;
677
678 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
679 n = kmalloc(size, GFP_KERNEL);
680 BUG_ON(!n);
681 memcpy(n, e820_table_kexec, size);
682 e820_table_kexec = n;
683
684 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
685 n = kmalloc(size, GFP_KERNEL);
686 BUG_ON(!n);
687 memcpy(n, e820_table_firmware, size);
688 e820_table_firmware = n;
689}
690
691/*
692 * Because of the small fixed size of struct boot_params, only the first
693 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
694 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
695 * struct setup_data, which is parsed here.
696 */
697void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
698{
699 int entries;
700 struct boot_e820_entry *extmap;
701 struct setup_data *sdata;
702
703 sdata = early_memremap(phys_addr, data_len);
704 entries = sdata->len / sizeof(*extmap);
705 extmap = (struct boot_e820_entry *)(sdata->data);
706
707 __append_e820_table(extmap, entries);
708 e820__update_table(e820_table);
709
710 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
711 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
712
713 early_memunmap(sdata, data_len);
714 pr_info("e820: extended physical RAM map:\n");
715 e820__print_table("extended");
716}
717
718/*
719 * Find the ranges of physical addresses that do not correspond to
720 * E820 RAM areas and register the corresponding pages as 'nosave' for
721 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
722 *
723 * This function requires the E820 map to be sorted and without any
724 * overlapping entries.
725 */
726void __init e820__register_nosave_regions(unsigned long limit_pfn)
727{
728 int i;
729 unsigned long pfn = 0;
730
731 for (i = 0; i < e820_table->nr_entries; i++) {
732 struct e820_entry *entry = &e820_table->entries[i];
733
734 if (pfn < PFN_UP(entry->addr))
735 register_nosave_region(pfn, PFN_UP(entry->addr));
736
737 pfn = PFN_DOWN(entry->addr + entry->size);
738
739 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
740 register_nosave_region(PFN_UP(entry->addr), pfn);
741
742 if (pfn >= limit_pfn)
743 break;
744 }
745}
746
747#ifdef CONFIG_ACPI
748/*
749 * Register ACPI NVS memory regions, so that we can save/restore them during
750 * hibernation and the subsequent resume:
751 */
752static int __init e820__register_nvs_regions(void)
753{
754 int i;
755
756 for (i = 0; i < e820_table->nr_entries; i++) {
757 struct e820_entry *entry = &e820_table->entries[i];
758
759 if (entry->type == E820_TYPE_NVS)
760 acpi_nvs_register(entry->addr, entry->size);
761 }
762
763 return 0;
764}
765core_initcall(e820__register_nvs_regions);
766#endif
767
768/*
769 * Allocate the requested number of bytes with the requsted alignment
770 * and return (the physical address) to the caller. Also register this
771 * range in the 'kexec' E820 table as a reserved range.
772 *
773 * This allows kexec to fake a new mptable, as if it came from the real
774 * system.
775 */
776u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
777{
778 u64 addr;
779
780 addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
781 if (addr) {
782 e820__range_update_kexec(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
783 pr_info("e820: update e820_table_kexec for e820__memblock_alloc_reserved()\n");
784 e820__update_table_kexec();
785 }
786
787 return addr;
788}
789
790#ifdef CONFIG_X86_32
791# ifdef CONFIG_X86_PAE
792# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
793# else
794# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
795# endif
796#else /* CONFIG_X86_32 */
797# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
798#endif
799
800/*
801 * Find the highest page frame number we have available
802 */
803static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
804{
805 int i;
806 unsigned long last_pfn = 0;
807 unsigned long max_arch_pfn = MAX_ARCH_PFN;
808
809 for (i = 0; i < e820_table->nr_entries; i++) {
810 struct e820_entry *entry = &e820_table->entries[i];
811 unsigned long start_pfn;
812 unsigned long end_pfn;
813
814 if (entry->type != type)
815 continue;
816
817 start_pfn = entry->addr >> PAGE_SHIFT;
818 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
819
820 if (start_pfn >= limit_pfn)
821 continue;
822 if (end_pfn > limit_pfn) {
823 last_pfn = limit_pfn;
824 break;
825 }
826 if (end_pfn > last_pfn)
827 last_pfn = end_pfn;
828 }
829
830 if (last_pfn > max_arch_pfn)
831 last_pfn = max_arch_pfn;
832
833 pr_info("e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
834 last_pfn, max_arch_pfn);
835 return last_pfn;
836}
837
838unsigned long __init e820__end_of_ram_pfn(void)
839{
840 return e820_end_pfn(MAX_ARCH_PFN, E820_TYPE_RAM);
841}
842
843unsigned long __init e820__end_of_low_ram_pfn(void)
844{
845 return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
846}
847
848static void __init early_panic(char *msg)
849{
850 early_printk(msg);
851 panic(msg);
852}
853
854static int userdef __initdata;
855
856/* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
857static int __init parse_memopt(char *p)
858{
859 u64 mem_size;
860
861 if (!p)
862 return -EINVAL;
863
864 if (!strcmp(p, "nopentium")) {
865#ifdef CONFIG_X86_32
866 setup_clear_cpu_cap(X86_FEATURE_PSE);
867 return 0;
868#else
869 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
870 return -EINVAL;
871#endif
872 }
873
874 userdef = 1;
875 mem_size = memparse(p, &p);
876
877 /* Don't remove all memory when getting "mem={invalid}" parameter: */
878 if (mem_size == 0)
879 return -EINVAL;
880
881 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
882
883 return 0;
884}
885early_param("mem", parse_memopt);
886
887static int __init parse_memmap_one(char *p)
888{
889 char *oldp;
890 u64 start_at, mem_size;
891
892 if (!p)
893 return -EINVAL;
894
895 if (!strncmp(p, "exactmap", 8)) {
896#ifdef CONFIG_CRASH_DUMP
897 /*
898 * If we are doing a crash dump, we still need to know
899 * the real memory size before the original memory map is
900 * reset.
901 */
902 saved_max_pfn = e820__end_of_ram_pfn();
903#endif
904 e820_table->nr_entries = 0;
905 userdef = 1;
906 return 0;
907 }
908
909 oldp = p;
910 mem_size = memparse(p, &p);
911 if (p == oldp)
912 return -EINVAL;
913
914 userdef = 1;
915 if (*p == '@') {
916 start_at = memparse(p+1, &p);
917 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
918 } else if (*p == '#') {
919 start_at = memparse(p+1, &p);
920 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
921 } else if (*p == '$') {
922 start_at = memparse(p+1, &p);
923 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
924 } else if (*p == '!') {
925 start_at = memparse(p+1, &p);
926 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
927 } else if (*p == '%') {
928 enum e820_type from = 0, to = 0;
929
930 start_at = memparse(p + 1, &p);
931 if (*p == '-')
932 from = simple_strtoull(p + 1, &p, 0);
933 if (*p == '+')
934 to = simple_strtoull(p + 1, &p, 0);
935 if (*p != '\0')
936 return -EINVAL;
937 if (from && to)
938 e820__range_update(start_at, mem_size, from, to);
939 else if (to)
940 e820__range_add(start_at, mem_size, to);
941 else if (from)
942 e820__range_remove(start_at, mem_size, from, 1);
943 else
944 e820__range_remove(start_at, mem_size, 0, 0);
945 } else {
946 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
947 }
948
949 return *p == '\0' ? 0 : -EINVAL;
950}
951
952static int __init parse_memmap_opt(char *str)
953{
954 while (str) {
955 char *k = strchr(str, ',');
956
957 if (k)
958 *k++ = 0;
959
960 parse_memmap_one(str);
961 str = k;
962 }
963
964 return 0;
965}
966early_param("memmap", parse_memmap_opt);
967
968/*
969 * Reserve all entries from the bootloader's extensible data nodes list,
970 * because if present we are going to use it later on to fetch e820
971 * entries from it:
972 */
973void __init e820__reserve_setup_data(void)
974{
975 struct setup_data *data;
976 u64 pa_data;
977
978 pa_data = boot_params.hdr.setup_data;
979 if (!pa_data)
980 return;
981
982 while (pa_data) {
983 data = early_memremap(pa_data, sizeof(*data));
984 e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
985 e820__range_update_kexec(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
986 pa_data = data->next;
987 early_memunmap(data, sizeof(*data));
988 }
989
990 e820__update_table(e820_table);
991 e820__update_table(e820_table_kexec);
992
993 pr_info("extended physical RAM map:\n");
994 e820__print_table("reserve setup_data");
995}
996
997/*
998 * Called after parse_early_param(), after early parameters (such as mem=)
999 * have been processed, in which case we already have an E820 table filled in
1000 * via the parameter callback function(s), but it's not sorted and printed yet:
1001 */
1002void __init e820__finish_early_params(void)
1003{
1004 if (userdef) {
1005 if (e820__update_table(e820_table) < 0)
1006 early_panic("Invalid user supplied memory map");
1007
1008 pr_info("e820: user-defined physical RAM map:\n");
1009 e820__print_table("user");
1010 }
1011}
1012
1013static const char *__init e820_type_to_string(struct e820_entry *entry)
1014{
1015 switch (entry->type) {
1016 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1017 case E820_TYPE_RAM: return "System RAM";
1018 case E820_TYPE_ACPI: return "ACPI Tables";
1019 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1020 case E820_TYPE_UNUSABLE: return "Unusable memory";
1021 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1022 case E820_TYPE_PMEM: return "Persistent Memory";
1023 case E820_TYPE_RESERVED: return "Reserved";
1024 default: return "Unknown E820 type";
1025 }
1026}
1027
1028static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1029{
1030 switch (entry->type) {
1031 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1032 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1033 case E820_TYPE_ACPI: /* Fall-through: */
1034 case E820_TYPE_NVS: /* Fall-through: */
1035 case E820_TYPE_UNUSABLE: /* Fall-through: */
1036 case E820_TYPE_PRAM: /* Fall-through: */
1037 case E820_TYPE_PMEM: /* Fall-through: */
1038 case E820_TYPE_RESERVED: /* Fall-through: */
1039 default: return IORESOURCE_MEM;
1040 }
1041}
1042
1043static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1044{
1045 switch (entry->type) {
1046 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1047 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1048 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1049 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1050 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1051 case E820_TYPE_RAM: /* Fall-through: */
1052 case E820_TYPE_UNUSABLE: /* Fall-through: */
1053 case E820_TYPE_RESERVED: /* Fall-through: */
1054 default: return IORES_DESC_NONE;
1055 }
1056}
1057
1058static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1059{
1060 /* this is the legacy bios/dos rom-shadow + mmio region */
1061 if (res->start < (1ULL<<20))
1062 return true;
1063
1064 /*
1065 * Treat persistent memory like device memory, i.e. reserve it
1066 * for exclusive use of a driver
1067 */
1068 switch (type) {
1069 case E820_TYPE_RESERVED:
1070 case E820_TYPE_PRAM:
1071 case E820_TYPE_PMEM:
1072 return false;
1073 case E820_TYPE_RESERVED_KERN:
1074 case E820_TYPE_RAM:
1075 case E820_TYPE_ACPI:
1076 case E820_TYPE_NVS:
1077 case E820_TYPE_UNUSABLE:
1078 default:
1079 return true;
1080 }
1081}
1082
1083/*
1084 * Mark E820 reserved areas as busy for the resource manager:
1085 */
1086
1087static struct resource __initdata *e820_res;
1088
1089void __init e820__reserve_resources(void)
1090{
1091 int i;
1092 struct resource *res;
1093 u64 end;
1094
1095 res = alloc_bootmem(sizeof(*res) * e820_table->nr_entries);
1096 e820_res = res;
1097
1098 for (i = 0; i < e820_table->nr_entries; i++) {
1099 struct e820_entry *entry = e820_table->entries + i;
1100
1101 end = entry->addr + entry->size - 1;
1102 if (end != (resource_size_t)end) {
1103 res++;
1104 continue;
1105 }
1106 res->start = entry->addr;
1107 res->end = end;
1108 res->name = e820_type_to_string(entry);
1109 res->flags = e820_type_to_iomem_type(entry);
1110 res->desc = e820_type_to_iores_desc(entry);
1111
1112 /*
1113 * Don't register the region that could be conflicted with
1114 * PCI device BAR resources and insert them later in
1115 * pcibios_resource_survey():
1116 */
1117 if (do_mark_busy(entry->type, res)) {
1118 res->flags |= IORESOURCE_BUSY;
1119 insert_resource(&iomem_resource, res);
1120 }
1121 res++;
1122 }
1123
1124 /* Expose the bootloader-provided memory layout to the sysfs. */
1125 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1126 struct e820_entry *entry = e820_table_firmware->entries + i;
1127
1128 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1129 }
1130}
1131
1132/*
1133 * How much should we pad the end of RAM, depending on where it is?
1134 */
1135static unsigned long __init ram_alignment(resource_size_t pos)
1136{
1137 unsigned long mb = pos >> 20;
1138
1139 /* To 64kB in the first megabyte */
1140 if (!mb)
1141 return 64*1024;
1142
1143 /* To 1MB in the first 16MB */
1144 if (mb < 16)
1145 return 1024*1024;
1146
1147 /* To 64MB for anything above that */
1148 return 64*1024*1024;
1149}
1150
1151#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1152
1153void __init e820__reserve_resources_late(void)
1154{
1155 int i;
1156 struct resource *res;
1157
1158 res = e820_res;
1159 for (i = 0; i < e820_table->nr_entries; i++) {
1160 if (!res->parent && res->end)
1161 insert_resource_expand_to_fit(&iomem_resource, res);
1162 res++;
1163 }
1164
1165 /*
1166 * Try to bump up RAM regions to reasonable boundaries, to
1167 * avoid stolen RAM:
1168 */
1169 for (i = 0; i < e820_table->nr_entries; i++) {
1170 struct e820_entry *entry = &e820_table->entries[i];
1171 u64 start, end;
1172
1173 if (entry->type != E820_TYPE_RAM)
1174 continue;
1175
1176 start = entry->addr + entry->size;
1177 end = round_up(start, ram_alignment(start)) - 1;
1178 if (end > MAX_RESOURCE_SIZE)
1179 end = MAX_RESOURCE_SIZE;
1180 if (start >= end)
1181 continue;
1182
1183 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1184 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1185 }
1186}
1187
1188/*
1189 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1190 */
1191char *__init e820__memory_setup_default(void)
1192{
1193 char *who = "BIOS-e820";
1194
1195 /*
1196 * Try to copy the BIOS-supplied E820-map.
1197 *
1198 * Otherwise fake a memory map; one section from 0k->640k,
1199 * the next section from 1mb->appropriate_mem_k
1200 */
1201 if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1202 u64 mem_size;
1203
1204 /* Compare results from other methods and take the one that gives more RAM: */
1205 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1206 mem_size = boot_params.screen_info.ext_mem_k;
1207 who = "BIOS-88";
1208 } else {
1209 mem_size = boot_params.alt_mem_k;
1210 who = "BIOS-e801";
1211 }
1212
1213 e820_table->nr_entries = 0;
1214 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1215 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1216 }
1217
1218 /* We just appended a lot of ranges, sanitize the table: */
1219 e820__update_table(e820_table);
1220
1221 return who;
1222}
1223
1224/*
1225 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1226 * E820 map - with an optional platform quirk available for virtual platforms
1227 * to override this method of boot environment processing:
1228 */
1229void __init e820__memory_setup(void)
1230{
1231 char *who;
1232
1233 /* This is a firmware interface ABI - make sure we don't break it: */
1234 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1235
1236 who = x86_init.resources.memory_setup();
1237
1238 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1239 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1240
1241 pr_info("e820: BIOS-provided physical RAM map:\n");
1242 e820__print_table(who);
1243}
1244
1245void __init e820__memblock_setup(void)
1246{
1247 int i;
1248 u64 end;
1249
1250 /*
1251 * The bootstrap memblock region count maximum is 128 entries
1252 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1253 * than that - so allow memblock resizing.
1254 *
1255 * This is safe, because this call happens pretty late during x86 setup,
1256 * so we know about reserved memory regions already. (This is important
1257 * so that memblock resizing does no stomp over reserved areas.)
1258 */
1259 memblock_allow_resize();
1260
1261 for (i = 0; i < e820_table->nr_entries; i++) {
1262 struct e820_entry *entry = &e820_table->entries[i];
1263
1264 end = entry->addr + entry->size;
1265 if (end != (resource_size_t)end)
1266 continue;
1267
1268 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1269 continue;
1270
1271 memblock_add(entry->addr, entry->size);
1272 }
1273
1274 /* Throw away partial pages: */
1275 memblock_trim_memory(PAGE_SIZE);
1276
1277 memblock_dump_all();
1278}
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
535u64 __init e820__range_update_table(struct e820_table *t, u64 start, u64 size,
536 enum e820_type old_type, enum e820_type new_type)
537{
538 return __e820__range_update(t, start, size, old_type, new_type);
539}
540
541/* Remove a range of memory from the E820 table: */
542u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
543{
544 int i;
545 u64 end;
546 u64 real_removed_size = 0;
547
548 if (size > (ULLONG_MAX - start))
549 size = ULLONG_MAX - start;
550
551 end = start + size;
552 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
553 if (check_type)
554 e820_print_type(old_type);
555 pr_cont("\n");
556
557 for (i = 0; i < e820_table->nr_entries; i++) {
558 struct e820_entry *entry = &e820_table->entries[i];
559 u64 final_start, final_end;
560 u64 entry_end;
561
562 if (check_type && entry->type != old_type)
563 continue;
564
565 entry_end = entry->addr + entry->size;
566
567 /* Completely covered? */
568 if (entry->addr >= start && entry_end <= end) {
569 real_removed_size += entry->size;
570 memset(entry, 0, sizeof(*entry));
571 continue;
572 }
573
574 /* Is the new range completely covered? */
575 if (entry->addr < start && entry_end > end) {
576 e820__range_add(end, entry_end - end, entry->type);
577 entry->size = start - entry->addr;
578 real_removed_size += size;
579 continue;
580 }
581
582 /* Partially covered: */
583 final_start = max(start, entry->addr);
584 final_end = min(end, entry_end);
585 if (final_start >= final_end)
586 continue;
587
588 real_removed_size += final_end - final_start;
589
590 /*
591 * Left range could be head or tail, so need to update
592 * the size first:
593 */
594 entry->size -= final_end - final_start;
595 if (entry->addr < final_start)
596 continue;
597
598 entry->addr = final_end;
599 }
600 return real_removed_size;
601}
602
603void __init e820__update_table_print(void)
604{
605 if (e820__update_table(e820_table))
606 return;
607
608 pr_info("modified physical RAM map:\n");
609 e820__print_table("modified");
610}
611
612static void __init e820__update_table_kexec(void)
613{
614 e820__update_table(e820_table_kexec);
615}
616
617#define MAX_GAP_END 0x100000000ull
618
619/*
620 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
621 */
622static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
623{
624 unsigned long long last = MAX_GAP_END;
625 int i = e820_table->nr_entries;
626 int found = 0;
627
628 while (--i >= 0) {
629 unsigned long long start = e820_table->entries[i].addr;
630 unsigned long long end = start + e820_table->entries[i].size;
631
632 /*
633 * Since "last" is at most 4GB, we know we'll
634 * fit in 32 bits if this condition is true:
635 */
636 if (last > end) {
637 unsigned long gap = last - end;
638
639 if (gap >= *gapsize) {
640 *gapsize = gap;
641 *gapstart = end;
642 found = 1;
643 }
644 }
645 if (start < last)
646 last = start;
647 }
648 return found;
649}
650
651/*
652 * Search for the biggest gap in the low 32 bits of the E820
653 * memory space. We pass this space to the PCI subsystem, so
654 * that it can assign MMIO resources for hotplug or
655 * unconfigured devices in.
656 *
657 * Hopefully the BIOS let enough space left.
658 */
659__init void e820__setup_pci_gap(void)
660{
661 unsigned long gapstart, gapsize;
662 int found;
663
664 gapsize = 0x400000;
665 found = e820_search_gap(&gapstart, &gapsize);
666
667 if (!found) {
668#ifdef CONFIG_X86_64
669 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
670 pr_err("Cannot find an available gap in the 32-bit address range\n");
671 pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
672#else
673 gapstart = 0x10000000;
674#endif
675 }
676
677 /*
678 * e820__reserve_resources_late() protects stolen RAM already:
679 */
680 pci_mem_start = gapstart;
681
682 pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
683 gapstart, gapstart + gapsize - 1);
684}
685
686/*
687 * Called late during init, in free_initmem().
688 *
689 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
690 *
691 * Copy them to a (usually much smaller) dynamically allocated area that is
692 * sized precisely after the number of e820 entries.
693 *
694 * This is done after we've performed all the fixes and tweaks to the tables.
695 * All functions which modify them are __init functions, which won't exist
696 * after free_initmem().
697 */
698__init void e820__reallocate_tables(void)
699{
700 struct e820_table *n;
701 int size;
702
703 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
704 n = kmemdup(e820_table, size, GFP_KERNEL);
705 BUG_ON(!n);
706 e820_table = n;
707
708 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
709 n = kmemdup(e820_table_kexec, size, GFP_KERNEL);
710 BUG_ON(!n);
711 e820_table_kexec = n;
712
713 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
714 n = kmemdup(e820_table_firmware, size, GFP_KERNEL);
715 BUG_ON(!n);
716 e820_table_firmware = n;
717}
718
719/*
720 * Because of the small fixed size of struct boot_params, only the first
721 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
722 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
723 * struct setup_data, which is parsed here.
724 */
725void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
726{
727 int entries;
728 struct boot_e820_entry *extmap;
729 struct setup_data *sdata;
730
731 sdata = early_memremap(phys_addr, data_len);
732 entries = sdata->len / sizeof(*extmap);
733 extmap = (struct boot_e820_entry *)(sdata->data);
734
735 __append_e820_table(extmap, entries);
736 e820__update_table(e820_table);
737
738 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
739 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
740
741 early_memunmap(sdata, data_len);
742 pr_info("extended physical RAM map:\n");
743 e820__print_table("extended");
744}
745
746/*
747 * Find the ranges of physical addresses that do not correspond to
748 * E820 RAM areas and register the corresponding pages as 'nosave' for
749 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
750 *
751 * This function requires the E820 map to be sorted and without any
752 * overlapping entries.
753 */
754void __init e820__register_nosave_regions(unsigned long limit_pfn)
755{
756 int i;
757 unsigned long pfn = 0;
758
759 for (i = 0; i < e820_table->nr_entries; i++) {
760 struct e820_entry *entry = &e820_table->entries[i];
761
762 if (pfn < PFN_UP(entry->addr))
763 register_nosave_region(pfn, PFN_UP(entry->addr));
764
765 pfn = PFN_DOWN(entry->addr + entry->size);
766
767 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
768 register_nosave_region(PFN_UP(entry->addr), pfn);
769
770 if (pfn >= limit_pfn)
771 break;
772 }
773}
774
775#ifdef CONFIG_ACPI
776/*
777 * Register ACPI NVS memory regions, so that we can save/restore them during
778 * hibernation and the subsequent resume:
779 */
780static int __init e820__register_nvs_regions(void)
781{
782 int i;
783
784 for (i = 0; i < e820_table->nr_entries; i++) {
785 struct e820_entry *entry = &e820_table->entries[i];
786
787 if (entry->type == E820_TYPE_NVS)
788 acpi_nvs_register(entry->addr, entry->size);
789 }
790
791 return 0;
792}
793core_initcall(e820__register_nvs_regions);
794#endif
795
796/*
797 * Allocate the requested number of bytes with the requested alignment
798 * and return (the physical address) to the caller. Also register this
799 * range in the 'kexec' E820 table as a reserved range.
800 *
801 * This allows kexec to fake a new mptable, as if it came from the real
802 * system.
803 */
804u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
805{
806 u64 addr;
807
808 addr = memblock_phys_alloc(size, align);
809 if (addr) {
810 e820__range_update_table(e820_table_kexec, addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
811 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
812 e820__update_table_kexec();
813 }
814
815 return addr;
816}
817
818#ifdef CONFIG_X86_32
819# ifdef CONFIG_X86_PAE
820# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
821# else
822# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
823# endif
824#else /* CONFIG_X86_32 */
825# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
826#endif
827
828/*
829 * Find the highest page frame number we have available
830 */
831static unsigned long __init e820__end_ram_pfn(unsigned long limit_pfn)
832{
833 int i;
834 unsigned long last_pfn = 0;
835 unsigned long max_arch_pfn = MAX_ARCH_PFN;
836
837 for (i = 0; i < e820_table->nr_entries; i++) {
838 struct e820_entry *entry = &e820_table->entries[i];
839 unsigned long start_pfn;
840 unsigned long end_pfn;
841
842 if (entry->type != E820_TYPE_RAM &&
843 entry->type != E820_TYPE_ACPI)
844 continue;
845
846 start_pfn = entry->addr >> PAGE_SHIFT;
847 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
848
849 if (start_pfn >= limit_pfn)
850 continue;
851 if (end_pfn > limit_pfn) {
852 last_pfn = limit_pfn;
853 break;
854 }
855 if (end_pfn > last_pfn)
856 last_pfn = end_pfn;
857 }
858
859 if (last_pfn > max_arch_pfn)
860 last_pfn = max_arch_pfn;
861
862 pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
863 last_pfn, max_arch_pfn);
864 return last_pfn;
865}
866
867unsigned long __init e820__end_of_ram_pfn(void)
868{
869 return e820__end_ram_pfn(MAX_ARCH_PFN);
870}
871
872unsigned long __init e820__end_of_low_ram_pfn(void)
873{
874 return e820__end_ram_pfn(1UL << (32 - PAGE_SHIFT));
875}
876
877static void __init early_panic(char *msg)
878{
879 early_printk(msg);
880 panic(msg);
881}
882
883static int userdef __initdata;
884
885/* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
886static int __init parse_memopt(char *p)
887{
888 u64 mem_size;
889
890 if (!p)
891 return -EINVAL;
892
893 if (!strcmp(p, "nopentium")) {
894#ifdef CONFIG_X86_32
895 setup_clear_cpu_cap(X86_FEATURE_PSE);
896 return 0;
897#else
898 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
899 return -EINVAL;
900#endif
901 }
902
903 userdef = 1;
904 mem_size = memparse(p, &p);
905
906 /* Don't remove all memory when getting "mem={invalid}" parameter: */
907 if (mem_size == 0)
908 return -EINVAL;
909
910 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
911
912#ifdef CONFIG_MEMORY_HOTPLUG
913 max_mem_size = mem_size;
914#endif
915
916 return 0;
917}
918early_param("mem", parse_memopt);
919
920static int __init parse_memmap_one(char *p)
921{
922 char *oldp;
923 u64 start_at, mem_size;
924
925 if (!p)
926 return -EINVAL;
927
928 if (!strncmp(p, "exactmap", 8)) {
929 e820_table->nr_entries = 0;
930 userdef = 1;
931 return 0;
932 }
933
934 oldp = p;
935 mem_size = memparse(p, &p);
936 if (p == oldp)
937 return -EINVAL;
938
939 userdef = 1;
940 if (*p == '@') {
941 start_at = memparse(p+1, &p);
942 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
943 } else if (*p == '#') {
944 start_at = memparse(p+1, &p);
945 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
946 } else if (*p == '$') {
947 start_at = memparse(p+1, &p);
948 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
949 } else if (*p == '!') {
950 start_at = memparse(p+1, &p);
951 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
952 } else if (*p == '%') {
953 enum e820_type from = 0, to = 0;
954
955 start_at = memparse(p + 1, &p);
956 if (*p == '-')
957 from = simple_strtoull(p + 1, &p, 0);
958 if (*p == '+')
959 to = simple_strtoull(p + 1, &p, 0);
960 if (*p != '\0')
961 return -EINVAL;
962 if (from && to)
963 e820__range_update(start_at, mem_size, from, to);
964 else if (to)
965 e820__range_add(start_at, mem_size, to);
966 else if (from)
967 e820__range_remove(start_at, mem_size, from, 1);
968 else
969 e820__range_remove(start_at, mem_size, 0, 0);
970 } else {
971 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
972 }
973
974 return *p == '\0' ? 0 : -EINVAL;
975}
976
977static int __init parse_memmap_opt(char *str)
978{
979 while (str) {
980 char *k = strchr(str, ',');
981
982 if (k)
983 *k++ = 0;
984
985 parse_memmap_one(str);
986 str = k;
987 }
988
989 return 0;
990}
991early_param("memmap", parse_memmap_opt);
992
993/*
994 * Reserve all entries from the bootloader's extensible data nodes list,
995 * because if present we are going to use it later on to fetch e820
996 * entries from it:
997 */
998void __init e820__reserve_setup_data(void)
999{
1000 struct setup_indirect *indirect;
1001 struct setup_data *data;
1002 u64 pa_data, pa_next;
1003 u32 len;
1004
1005 pa_data = boot_params.hdr.setup_data;
1006 if (!pa_data)
1007 return;
1008
1009 while (pa_data) {
1010 data = early_memremap(pa_data, sizeof(*data));
1011 if (!data) {
1012 pr_warn("e820: failed to memremap setup_data entry\n");
1013 return;
1014 }
1015
1016 len = sizeof(*data);
1017 pa_next = data->next;
1018
1019 e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1020
1021 if (data->type == SETUP_INDIRECT) {
1022 len += data->len;
1023 early_memunmap(data, sizeof(*data));
1024 data = early_memremap(pa_data, len);
1025 if (!data) {
1026 pr_warn("e820: failed to memremap indirect setup_data\n");
1027 return;
1028 }
1029
1030 indirect = (struct setup_indirect *)data->data;
1031
1032 if (indirect->type != SETUP_INDIRECT)
1033 e820__range_update(indirect->addr, indirect->len,
1034 E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1035 }
1036
1037 pa_data = pa_next;
1038 early_memunmap(data, len);
1039 }
1040
1041 e820__update_table(e820_table);
1042
1043 pr_info("extended physical RAM map:\n");
1044 e820__print_table("reserve setup_data");
1045}
1046
1047/*
1048 * Called after parse_early_param(), after early parameters (such as mem=)
1049 * have been processed, in which case we already have an E820 table filled in
1050 * via the parameter callback function(s), but it's not sorted and printed yet:
1051 */
1052void __init e820__finish_early_params(void)
1053{
1054 if (userdef) {
1055 if (e820__update_table(e820_table) < 0)
1056 early_panic("Invalid user supplied memory map");
1057
1058 pr_info("user-defined physical RAM map:\n");
1059 e820__print_table("user");
1060 }
1061}
1062
1063static const char *__init e820_type_to_string(struct e820_entry *entry)
1064{
1065 switch (entry->type) {
1066 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1067 case E820_TYPE_RAM: return "System RAM";
1068 case E820_TYPE_ACPI: return "ACPI Tables";
1069 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1070 case E820_TYPE_UNUSABLE: return "Unusable memory";
1071 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1072 case E820_TYPE_PMEM: return "Persistent Memory";
1073 case E820_TYPE_RESERVED: return "Reserved";
1074 case E820_TYPE_SOFT_RESERVED: return "Soft Reserved";
1075 default: return "Unknown E820 type";
1076 }
1077}
1078
1079static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1080{
1081 switch (entry->type) {
1082 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1083 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1084 case E820_TYPE_ACPI: /* Fall-through: */
1085 case E820_TYPE_NVS: /* Fall-through: */
1086 case E820_TYPE_UNUSABLE: /* Fall-through: */
1087 case E820_TYPE_PRAM: /* Fall-through: */
1088 case E820_TYPE_PMEM: /* Fall-through: */
1089 case E820_TYPE_RESERVED: /* Fall-through: */
1090 case E820_TYPE_SOFT_RESERVED: /* Fall-through: */
1091 default: return IORESOURCE_MEM;
1092 }
1093}
1094
1095static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1096{
1097 switch (entry->type) {
1098 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1099 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1100 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1101 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1102 case E820_TYPE_RESERVED: return IORES_DESC_RESERVED;
1103 case E820_TYPE_SOFT_RESERVED: return IORES_DESC_SOFT_RESERVED;
1104 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1105 case E820_TYPE_RAM: /* Fall-through: */
1106 case E820_TYPE_UNUSABLE: /* Fall-through: */
1107 default: return IORES_DESC_NONE;
1108 }
1109}
1110
1111static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1112{
1113 /* this is the legacy bios/dos rom-shadow + mmio region */
1114 if (res->start < (1ULL<<20))
1115 return true;
1116
1117 /*
1118 * Treat persistent memory and other special memory ranges like
1119 * device memory, i.e. reserve it for exclusive use of a driver
1120 */
1121 switch (type) {
1122 case E820_TYPE_RESERVED:
1123 case E820_TYPE_SOFT_RESERVED:
1124 case E820_TYPE_PRAM:
1125 case E820_TYPE_PMEM:
1126 return false;
1127 case E820_TYPE_RESERVED_KERN:
1128 case E820_TYPE_RAM:
1129 case E820_TYPE_ACPI:
1130 case E820_TYPE_NVS:
1131 case E820_TYPE_UNUSABLE:
1132 default:
1133 return true;
1134 }
1135}
1136
1137/*
1138 * Mark E820 reserved areas as busy for the resource manager:
1139 */
1140
1141static struct resource __initdata *e820_res;
1142
1143void __init e820__reserve_resources(void)
1144{
1145 int i;
1146 struct resource *res;
1147 u64 end;
1148
1149 res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1150 SMP_CACHE_BYTES);
1151 if (!res)
1152 panic("%s: Failed to allocate %zu bytes\n", __func__,
1153 sizeof(*res) * e820_table->nr_entries);
1154 e820_res = res;
1155
1156 for (i = 0; i < e820_table->nr_entries; i++) {
1157 struct e820_entry *entry = e820_table->entries + i;
1158
1159 end = entry->addr + entry->size - 1;
1160 if (end != (resource_size_t)end) {
1161 res++;
1162 continue;
1163 }
1164 res->start = entry->addr;
1165 res->end = end;
1166 res->name = e820_type_to_string(entry);
1167 res->flags = e820_type_to_iomem_type(entry);
1168 res->desc = e820_type_to_iores_desc(entry);
1169
1170 /*
1171 * Don't register the region that could be conflicted with
1172 * PCI device BAR resources and insert them later in
1173 * pcibios_resource_survey():
1174 */
1175 if (do_mark_busy(entry->type, res)) {
1176 res->flags |= IORESOURCE_BUSY;
1177 insert_resource(&iomem_resource, res);
1178 }
1179 res++;
1180 }
1181
1182 /* Expose the bootloader-provided memory layout to the sysfs. */
1183 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1184 struct e820_entry *entry = e820_table_firmware->entries + i;
1185
1186 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1187 }
1188}
1189
1190/*
1191 * How much should we pad the end of RAM, depending on where it is?
1192 */
1193static unsigned long __init ram_alignment(resource_size_t pos)
1194{
1195 unsigned long mb = pos >> 20;
1196
1197 /* To 64kB in the first megabyte */
1198 if (!mb)
1199 return 64*1024;
1200
1201 /* To 1MB in the first 16MB */
1202 if (mb < 16)
1203 return 1024*1024;
1204
1205 /* To 64MB for anything above that */
1206 return 64*1024*1024;
1207}
1208
1209#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1210
1211void __init e820__reserve_resources_late(void)
1212{
1213 int i;
1214 struct resource *res;
1215
1216 res = e820_res;
1217 for (i = 0; i < e820_table->nr_entries; i++) {
1218 if (!res->parent && res->end)
1219 insert_resource_expand_to_fit(&iomem_resource, res);
1220 res++;
1221 }
1222
1223 /*
1224 * Try to bump up RAM regions to reasonable boundaries, to
1225 * avoid stolen RAM:
1226 */
1227 for (i = 0; i < e820_table->nr_entries; i++) {
1228 struct e820_entry *entry = &e820_table->entries[i];
1229 u64 start, end;
1230
1231 if (entry->type != E820_TYPE_RAM)
1232 continue;
1233
1234 start = entry->addr + entry->size;
1235 end = round_up(start, ram_alignment(start)) - 1;
1236 if (end > MAX_RESOURCE_SIZE)
1237 end = MAX_RESOURCE_SIZE;
1238 if (start >= end)
1239 continue;
1240
1241 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1242 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1243 }
1244}
1245
1246/*
1247 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1248 */
1249char *__init e820__memory_setup_default(void)
1250{
1251 char *who = "BIOS-e820";
1252
1253 /*
1254 * Try to copy the BIOS-supplied E820-map.
1255 *
1256 * Otherwise fake a memory map; one section from 0k->640k,
1257 * the next section from 1mb->appropriate_mem_k
1258 */
1259 if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1260 u64 mem_size;
1261
1262 /* Compare results from other methods and take the one that gives more RAM: */
1263 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1264 mem_size = boot_params.screen_info.ext_mem_k;
1265 who = "BIOS-88";
1266 } else {
1267 mem_size = boot_params.alt_mem_k;
1268 who = "BIOS-e801";
1269 }
1270
1271 e820_table->nr_entries = 0;
1272 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1273 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1274 }
1275
1276 /* We just appended a lot of ranges, sanitize the table: */
1277 e820__update_table(e820_table);
1278
1279 return who;
1280}
1281
1282/*
1283 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1284 * E820 map - with an optional platform quirk available for virtual platforms
1285 * to override this method of boot environment processing:
1286 */
1287void __init e820__memory_setup(void)
1288{
1289 char *who;
1290
1291 /* This is a firmware interface ABI - make sure we don't break it: */
1292 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1293
1294 who = x86_init.resources.memory_setup();
1295
1296 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1297 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1298
1299 pr_info("BIOS-provided physical RAM map:\n");
1300 e820__print_table(who);
1301}
1302
1303void __init e820__memblock_setup(void)
1304{
1305 int i;
1306 u64 end;
1307
1308 /*
1309 * The bootstrap memblock region count maximum is 128 entries
1310 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1311 * than that - so allow memblock resizing.
1312 *
1313 * This is safe, because this call happens pretty late during x86 setup,
1314 * so we know about reserved memory regions already. (This is important
1315 * so that memblock resizing does no stomp over reserved areas.)
1316 */
1317 memblock_allow_resize();
1318
1319 for (i = 0; i < e820_table->nr_entries; i++) {
1320 struct e820_entry *entry = &e820_table->entries[i];
1321
1322 end = entry->addr + entry->size;
1323 if (end != (resource_size_t)end)
1324 continue;
1325
1326 if (entry->type == E820_TYPE_SOFT_RESERVED)
1327 memblock_reserve(entry->addr, entry->size);
1328
1329 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1330 continue;
1331
1332 memblock_add(entry->addr, entry->size);
1333 }
1334
1335 /* Throw away partial pages: */
1336 memblock_trim_memory(PAGE_SIZE);
1337
1338 memblock_dump_all();
1339}