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/*
2 * Handle the memory map.
3 * The functions here do the job until bootmem takes over.
4 *
5 * Getting sanitize_e820_map() in sync with i386 version by applying change:
6 * - Provisions for empty E820 memory regions (reported by certain BIOSes).
7 * Alex Achenbach <xela@slit.de>, December 2002.
8 * Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
9 *
10 */
11#include <linux/kernel.h>
12#include <linux/types.h>
13#include <linux/init.h>
14#include <linux/crash_dump.h>
15#include <linux/export.h>
16#include <linux/bootmem.h>
17#include <linux/pfn.h>
18#include <linux/suspend.h>
19#include <linux/acpi.h>
20#include <linux/firmware-map.h>
21#include <linux/memblock.h>
22#include <linux/sort.h>
23
24#include <asm/e820.h>
25#include <asm/proto.h>
26#include <asm/setup.h>
27#include <asm/cpufeature.h>
28
29/*
30 * The e820 map is the map that gets modified e.g. with command line parameters
31 * and that is also registered with modifications in the kernel resource tree
32 * with the iomem_resource as parent.
33 *
34 * The e820_saved is directly saved after the BIOS-provided memory map is
35 * copied. It doesn't get modified afterwards. It's registered for the
36 * /sys/firmware/memmap interface.
37 *
38 * That memory map is not modified and is used as base for kexec. The kexec'd
39 * kernel should get the same memory map as the firmware provides. Then the
40 * user can e.g. boot the original kernel with mem=1G while still booting the
41 * next kernel with full memory.
42 */
43struct e820map e820;
44struct e820map e820_saved;
45
46/* For PCI or other memory-mapped resources */
47unsigned long pci_mem_start = 0xaeedbabe;
48#ifdef CONFIG_PCI
49EXPORT_SYMBOL(pci_mem_start);
50#endif
51
52/*
53 * This function checks if any part of the range <start,end> is mapped
54 * with type.
55 */
56int
57e820_any_mapped(u64 start, u64 end, unsigned type)
58{
59 int i;
60
61 for (i = 0; i < e820.nr_map; i++) {
62 struct e820entry *ei = &e820.map[i];
63
64 if (type && ei->type != type)
65 continue;
66 if (ei->addr >= end || ei->addr + ei->size <= start)
67 continue;
68 return 1;
69 }
70 return 0;
71}
72EXPORT_SYMBOL_GPL(e820_any_mapped);
73
74/*
75 * This function checks if the entire range <start,end> is mapped with type.
76 *
77 * Note: this function only works correct if the e820 table is sorted and
78 * not-overlapping, which is the case
79 */
80int __init e820_all_mapped(u64 start, u64 end, unsigned type)
81{
82 int i;
83
84 for (i = 0; i < e820.nr_map; i++) {
85 struct e820entry *ei = &e820.map[i];
86
87 if (type && ei->type != type)
88 continue;
89 /* is the region (part) in overlap with the current region ?*/
90 if (ei->addr >= end || ei->addr + ei->size <= start)
91 continue;
92
93 /* if the region is at the beginning of <start,end> we move
94 * start to the end of the region since it's ok until there
95 */
96 if (ei->addr <= start)
97 start = ei->addr + ei->size;
98 /*
99 * if start is now at or beyond end, we're done, full
100 * coverage
101 */
102 if (start >= end)
103 return 1;
104 }
105 return 0;
106}
107
108/*
109 * Add a memory region to the kernel e820 map.
110 */
111static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
112 int type)
113{
114 int x = e820x->nr_map;
115
116 if (x >= ARRAY_SIZE(e820x->map)) {
117 printk(KERN_ERR "e820: too many entries; ignoring [mem %#010llx-%#010llx]\n",
118 (unsigned long long) start,
119 (unsigned long long) (start + size - 1));
120 return;
121 }
122
123 e820x->map[x].addr = start;
124 e820x->map[x].size = size;
125 e820x->map[x].type = type;
126 e820x->nr_map++;
127}
128
129void __init e820_add_region(u64 start, u64 size, int type)
130{
131 __e820_add_region(&e820, start, size, type);
132}
133
134static void __init e820_print_type(u32 type)
135{
136 switch (type) {
137 case E820_RAM:
138 case E820_RESERVED_KERN:
139 printk(KERN_CONT "usable");
140 break;
141 case E820_RESERVED:
142 printk(KERN_CONT "reserved");
143 break;
144 case E820_ACPI:
145 printk(KERN_CONT "ACPI data");
146 break;
147 case E820_NVS:
148 printk(KERN_CONT "ACPI NVS");
149 break;
150 case E820_UNUSABLE:
151 printk(KERN_CONT "unusable");
152 break;
153 case E820_PMEM:
154 case E820_PRAM:
155 printk(KERN_CONT "persistent (type %u)", type);
156 break;
157 default:
158 printk(KERN_CONT "type %u", type);
159 break;
160 }
161}
162
163void __init e820_print_map(char *who)
164{
165 int i;
166
167 for (i = 0; i < e820.nr_map; i++) {
168 printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who,
169 (unsigned long long) e820.map[i].addr,
170 (unsigned long long)
171 (e820.map[i].addr + e820.map[i].size - 1));
172 e820_print_type(e820.map[i].type);
173 printk(KERN_CONT "\n");
174 }
175}
176
177/*
178 * Sanitize the BIOS e820 map.
179 *
180 * Some e820 responses include overlapping entries. The following
181 * replaces the original e820 map with a new one, removing overlaps,
182 * and resolving conflicting memory types in favor of highest
183 * numbered type.
184 *
185 * The input parameter biosmap points to an array of 'struct
186 * e820entry' which on entry has elements in the range [0, *pnr_map)
187 * valid, and which has space for up to max_nr_map entries.
188 * On return, the resulting sanitized e820 map entries will be in
189 * overwritten in the same location, starting at biosmap.
190 *
191 * The integer pointed to by pnr_map must be valid on entry (the
192 * current number of valid entries located at biosmap). If the
193 * sanitizing succeeds the *pnr_map will be updated with the new
194 * number of valid entries (something no more than max_nr_map).
195 *
196 * The return value from sanitize_e820_map() is zero if it
197 * successfully 'sanitized' the map entries passed in, and is -1
198 * if it did nothing, which can happen if either of (1) it was
199 * only passed one map entry, or (2) any of the input map entries
200 * were invalid (start + size < start, meaning that the size was
201 * so big the described memory range wrapped around through zero.)
202 *
203 * Visually we're performing the following
204 * (1,2,3,4 = memory types)...
205 *
206 * Sample memory map (w/overlaps):
207 * ____22__________________
208 * ______________________4_
209 * ____1111________________
210 * _44_____________________
211 * 11111111________________
212 * ____________________33__
213 * ___________44___________
214 * __________33333_________
215 * ______________22________
216 * ___________________2222_
217 * _________111111111______
218 * _____________________11_
219 * _________________4______
220 *
221 * Sanitized equivalent (no overlap):
222 * 1_______________________
223 * _44_____________________
224 * ___1____________________
225 * ____22__________________
226 * ______11________________
227 * _________1______________
228 * __________3_____________
229 * ___________44___________
230 * _____________33_________
231 * _______________2________
232 * ________________1_______
233 * _________________4______
234 * ___________________2____
235 * ____________________33__
236 * ______________________4_
237 */
238struct change_member {
239 struct e820entry *pbios; /* pointer to original bios entry */
240 unsigned long long addr; /* address for this change point */
241};
242
243static int __init cpcompare(const void *a, const void *b)
244{
245 struct change_member * const *app = a, * const *bpp = b;
246 const struct change_member *ap = *app, *bp = *bpp;
247
248 /*
249 * Inputs are pointers to two elements of change_point[]. If their
250 * addresses are unequal, their difference dominates. If the addresses
251 * are equal, then consider one that represents the end of its region
252 * to be greater than one that does not.
253 */
254 if (ap->addr != bp->addr)
255 return ap->addr > bp->addr ? 1 : -1;
256
257 return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr);
258}
259
260int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
261 u32 *pnr_map)
262{
263 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
264 static struct change_member *change_point[2*E820_X_MAX] __initdata;
265 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
266 static struct e820entry new_bios[E820_X_MAX] __initdata;
267 unsigned long current_type, last_type;
268 unsigned long long last_addr;
269 int chgidx;
270 int overlap_entries;
271 int new_bios_entry;
272 int old_nr, new_nr, chg_nr;
273 int i;
274
275 /* if there's only one memory region, don't bother */
276 if (*pnr_map < 2)
277 return -1;
278
279 old_nr = *pnr_map;
280 BUG_ON(old_nr > max_nr_map);
281
282 /* bail out if we find any unreasonable addresses in bios map */
283 for (i = 0; i < old_nr; i++)
284 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
285 return -1;
286
287 /* create pointers for initial change-point information (for sorting) */
288 for (i = 0; i < 2 * old_nr; i++)
289 change_point[i] = &change_point_list[i];
290
291 /* record all known change-points (starting and ending addresses),
292 omitting those that are for empty memory regions */
293 chgidx = 0;
294 for (i = 0; i < old_nr; i++) {
295 if (biosmap[i].size != 0) {
296 change_point[chgidx]->addr = biosmap[i].addr;
297 change_point[chgidx++]->pbios = &biosmap[i];
298 change_point[chgidx]->addr = biosmap[i].addr +
299 biosmap[i].size;
300 change_point[chgidx++]->pbios = &biosmap[i];
301 }
302 }
303 chg_nr = chgidx;
304
305 /* sort change-point list by memory addresses (low -> high) */
306 sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL);
307
308 /* create a new bios memory map, removing overlaps */
309 overlap_entries = 0; /* number of entries in the overlap table */
310 new_bios_entry = 0; /* index for creating new bios map entries */
311 last_type = 0; /* start with undefined memory type */
312 last_addr = 0; /* start with 0 as last starting address */
313
314 /* loop through change-points, determining affect on the new bios map */
315 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
316 /* keep track of all overlapping bios entries */
317 if (change_point[chgidx]->addr ==
318 change_point[chgidx]->pbios->addr) {
319 /*
320 * add map entry to overlap list (> 1 entry
321 * implies an overlap)
322 */
323 overlap_list[overlap_entries++] =
324 change_point[chgidx]->pbios;
325 } else {
326 /*
327 * remove entry from list (order independent,
328 * so swap with last)
329 */
330 for (i = 0; i < overlap_entries; i++) {
331 if (overlap_list[i] ==
332 change_point[chgidx]->pbios)
333 overlap_list[i] =
334 overlap_list[overlap_entries-1];
335 }
336 overlap_entries--;
337 }
338 /*
339 * if there are overlapping entries, decide which
340 * "type" to use (larger value takes precedence --
341 * 1=usable, 2,3,4,4+=unusable)
342 */
343 current_type = 0;
344 for (i = 0; i < overlap_entries; i++)
345 if (overlap_list[i]->type > current_type)
346 current_type = overlap_list[i]->type;
347 /*
348 * continue building up new bios map based on this
349 * information
350 */
351 if (current_type != last_type || current_type == E820_PRAM) {
352 if (last_type != 0) {
353 new_bios[new_bios_entry].size =
354 change_point[chgidx]->addr - last_addr;
355 /*
356 * move forward only if the new size
357 * was non-zero
358 */
359 if (new_bios[new_bios_entry].size != 0)
360 /*
361 * no more space left for new
362 * bios entries ?
363 */
364 if (++new_bios_entry >= max_nr_map)
365 break;
366 }
367 if (current_type != 0) {
368 new_bios[new_bios_entry].addr =
369 change_point[chgidx]->addr;
370 new_bios[new_bios_entry].type = current_type;
371 last_addr = change_point[chgidx]->addr;
372 }
373 last_type = current_type;
374 }
375 }
376 /* retain count for new bios entries */
377 new_nr = new_bios_entry;
378
379 /* copy new bios mapping into original location */
380 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
381 *pnr_map = new_nr;
382
383 return 0;
384}
385
386static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
387{
388 while (nr_map) {
389 u64 start = biosmap->addr;
390 u64 size = biosmap->size;
391 u64 end = start + size;
392 u32 type = biosmap->type;
393
394 /* Overflow in 64 bits? Ignore the memory map. */
395 if (start > end)
396 return -1;
397
398 e820_add_region(start, size, type);
399
400 biosmap++;
401 nr_map--;
402 }
403 return 0;
404}
405
406/*
407 * Copy the BIOS e820 map into a safe place.
408 *
409 * Sanity-check it while we're at it..
410 *
411 * If we're lucky and live on a modern system, the setup code
412 * will have given us a memory map that we can use to properly
413 * set up memory. If we aren't, we'll fake a memory map.
414 */
415static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
416{
417 /* Only one memory region (or negative)? Ignore it */
418 if (nr_map < 2)
419 return -1;
420
421 return __append_e820_map(biosmap, nr_map);
422}
423
424static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
425 u64 size, unsigned old_type,
426 unsigned new_type)
427{
428 u64 end;
429 unsigned int i;
430 u64 real_updated_size = 0;
431
432 BUG_ON(old_type == new_type);
433
434 if (size > (ULLONG_MAX - start))
435 size = ULLONG_MAX - start;
436
437 end = start + size;
438 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ",
439 (unsigned long long) start, (unsigned long long) (end - 1));
440 e820_print_type(old_type);
441 printk(KERN_CONT " ==> ");
442 e820_print_type(new_type);
443 printk(KERN_CONT "\n");
444
445 for (i = 0; i < e820x->nr_map; i++) {
446 struct e820entry *ei = &e820x->map[i];
447 u64 final_start, final_end;
448 u64 ei_end;
449
450 if (ei->type != old_type)
451 continue;
452
453 ei_end = ei->addr + ei->size;
454 /* totally covered by new range? */
455 if (ei->addr >= start && ei_end <= end) {
456 ei->type = new_type;
457 real_updated_size += ei->size;
458 continue;
459 }
460
461 /* new range is totally covered? */
462 if (ei->addr < start && ei_end > end) {
463 __e820_add_region(e820x, start, size, new_type);
464 __e820_add_region(e820x, end, ei_end - end, ei->type);
465 ei->size = start - ei->addr;
466 real_updated_size += size;
467 continue;
468 }
469
470 /* partially covered */
471 final_start = max(start, ei->addr);
472 final_end = min(end, ei_end);
473 if (final_start >= final_end)
474 continue;
475
476 __e820_add_region(e820x, final_start, final_end - final_start,
477 new_type);
478
479 real_updated_size += final_end - final_start;
480
481 /*
482 * left range could be head or tail, so need to update
483 * size at first.
484 */
485 ei->size -= final_end - final_start;
486 if (ei->addr < final_start)
487 continue;
488 ei->addr = final_end;
489 }
490 return real_updated_size;
491}
492
493u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
494 unsigned new_type)
495{
496 return __e820_update_range(&e820, start, size, old_type, new_type);
497}
498
499static u64 __init e820_update_range_saved(u64 start, u64 size,
500 unsigned old_type, unsigned new_type)
501{
502 return __e820_update_range(&e820_saved, start, size, old_type,
503 new_type);
504}
505
506/* make e820 not cover the range */
507u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
508 int checktype)
509{
510 int i;
511 u64 end;
512 u64 real_removed_size = 0;
513
514 if (size > (ULLONG_MAX - start))
515 size = ULLONG_MAX - start;
516
517 end = start + size;
518 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ",
519 (unsigned long long) start, (unsigned long long) (end - 1));
520 if (checktype)
521 e820_print_type(old_type);
522 printk(KERN_CONT "\n");
523
524 for (i = 0; i < e820.nr_map; i++) {
525 struct e820entry *ei = &e820.map[i];
526 u64 final_start, final_end;
527 u64 ei_end;
528
529 if (checktype && ei->type != old_type)
530 continue;
531
532 ei_end = ei->addr + ei->size;
533 /* totally covered? */
534 if (ei->addr >= start && ei_end <= end) {
535 real_removed_size += ei->size;
536 memset(ei, 0, sizeof(struct e820entry));
537 continue;
538 }
539
540 /* new range is totally covered? */
541 if (ei->addr < start && ei_end > end) {
542 e820_add_region(end, ei_end - end, ei->type);
543 ei->size = start - ei->addr;
544 real_removed_size += size;
545 continue;
546 }
547
548 /* partially covered */
549 final_start = max(start, ei->addr);
550 final_end = min(end, ei_end);
551 if (final_start >= final_end)
552 continue;
553 real_removed_size += final_end - final_start;
554
555 /*
556 * left range could be head or tail, so need to update
557 * size at first.
558 */
559 ei->size -= final_end - final_start;
560 if (ei->addr < final_start)
561 continue;
562 ei->addr = final_end;
563 }
564 return real_removed_size;
565}
566
567void __init update_e820(void)
568{
569 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map))
570 return;
571 printk(KERN_INFO "e820: modified physical RAM map:\n");
572 e820_print_map("modified");
573}
574static void __init update_e820_saved(void)
575{
576 sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map),
577 &e820_saved.nr_map);
578}
579#define MAX_GAP_END 0x100000000ull
580/*
581 * Search for a gap in the e820 memory space from start_addr to end_addr.
582 */
583__init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
584 unsigned long start_addr, unsigned long long end_addr)
585{
586 unsigned long long last;
587 int i = e820.nr_map;
588 int found = 0;
589
590 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
591
592 while (--i >= 0) {
593 unsigned long long start = e820.map[i].addr;
594 unsigned long long end = start + e820.map[i].size;
595
596 if (end < start_addr)
597 continue;
598
599 /*
600 * Since "last" is at most 4GB, we know we'll
601 * fit in 32 bits if this condition is true
602 */
603 if (last > end) {
604 unsigned long gap = last - end;
605
606 if (gap >= *gapsize) {
607 *gapsize = gap;
608 *gapstart = end;
609 found = 1;
610 }
611 }
612 if (start < last)
613 last = start;
614 }
615 return found;
616}
617
618/*
619 * Search for the biggest gap in the low 32 bits of the e820
620 * memory space. We pass this space to PCI to assign MMIO resources
621 * for hotplug or unconfigured devices in.
622 * Hopefully the BIOS let enough space left.
623 */
624__init void e820_setup_gap(void)
625{
626 unsigned long gapstart, gapsize;
627 int found;
628
629 gapstart = 0x10000000;
630 gapsize = 0x400000;
631 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
632
633#ifdef CONFIG_X86_64
634 if (!found) {
635 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
636 printk(KERN_ERR
637 "e820: cannot find a gap in the 32bit address range\n"
638 "e820: PCI devices with unassigned 32bit BARs may break!\n");
639 }
640#endif
641
642 /*
643 * e820_reserve_resources_late protect stolen RAM already
644 */
645 pci_mem_start = gapstart;
646
647 printk(KERN_INFO
648 "e820: [mem %#010lx-%#010lx] available for PCI devices\n",
649 gapstart, gapstart + gapsize - 1);
650}
651
652/**
653 * Because of the size limitation of struct boot_params, only first
654 * 128 E820 memory entries are passed to kernel via
655 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
656 * linked list of struct setup_data, which is parsed here.
657 */
658void __init parse_e820_ext(u64 phys_addr, u32 data_len)
659{
660 int entries;
661 struct e820entry *extmap;
662 struct setup_data *sdata;
663
664 sdata = early_memremap(phys_addr, data_len);
665 entries = sdata->len / sizeof(struct e820entry);
666 extmap = (struct e820entry *)(sdata->data);
667 __append_e820_map(extmap, entries);
668 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
669 early_memunmap(sdata, data_len);
670 printk(KERN_INFO "e820: extended physical RAM map:\n");
671 e820_print_map("extended");
672}
673
674#if defined(CONFIG_X86_64) || \
675 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
676/**
677 * Find the ranges of physical addresses that do not correspond to
678 * e820 RAM areas and mark the corresponding pages as nosave for
679 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
680 *
681 * This function requires the e820 map to be sorted and without any
682 * overlapping entries.
683 */
684void __init e820_mark_nosave_regions(unsigned long limit_pfn)
685{
686 int i;
687 unsigned long pfn = 0;
688
689 for (i = 0; i < e820.nr_map; i++) {
690 struct e820entry *ei = &e820.map[i];
691
692 if (pfn < PFN_UP(ei->addr))
693 register_nosave_region(pfn, PFN_UP(ei->addr));
694
695 pfn = PFN_DOWN(ei->addr + ei->size);
696
697 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
698 register_nosave_region(PFN_UP(ei->addr), pfn);
699
700 if (pfn >= limit_pfn)
701 break;
702 }
703}
704#endif
705
706#ifdef CONFIG_ACPI
707/**
708 * Mark ACPI NVS memory region, so that we can save/restore it during
709 * hibernation and the subsequent resume.
710 */
711static int __init e820_mark_nvs_memory(void)
712{
713 int i;
714
715 for (i = 0; i < e820.nr_map; i++) {
716 struct e820entry *ei = &e820.map[i];
717
718 if (ei->type == E820_NVS)
719 acpi_nvs_register(ei->addr, ei->size);
720 }
721
722 return 0;
723}
724core_initcall(e820_mark_nvs_memory);
725#endif
726
727/*
728 * pre allocated 4k and reserved it in memblock and e820_saved
729 */
730u64 __init early_reserve_e820(u64 size, u64 align)
731{
732 u64 addr;
733
734 addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
735 if (addr) {
736 e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED);
737 printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n");
738 update_e820_saved();
739 }
740
741 return addr;
742}
743
744#ifdef CONFIG_X86_32
745# ifdef CONFIG_X86_PAE
746# define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
747# else
748# define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
749# endif
750#else /* CONFIG_X86_32 */
751# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
752#endif
753
754/*
755 * Find the highest page frame number we have available
756 */
757static unsigned long __init e820_end_pfn(unsigned long limit_pfn)
758{
759 int i;
760 unsigned long last_pfn = 0;
761 unsigned long max_arch_pfn = MAX_ARCH_PFN;
762
763 for (i = 0; i < e820.nr_map; i++) {
764 struct e820entry *ei = &e820.map[i];
765 unsigned long start_pfn;
766 unsigned long end_pfn;
767
768 /*
769 * Persistent memory is accounted as ram for purposes of
770 * establishing max_pfn and mem_map.
771 */
772 if (ei->type != E820_RAM && ei->type != E820_PRAM)
773 continue;
774
775 start_pfn = ei->addr >> PAGE_SHIFT;
776 end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
777
778 if (start_pfn >= limit_pfn)
779 continue;
780 if (end_pfn > limit_pfn) {
781 last_pfn = limit_pfn;
782 break;
783 }
784 if (end_pfn > last_pfn)
785 last_pfn = end_pfn;
786 }
787
788 if (last_pfn > max_arch_pfn)
789 last_pfn = max_arch_pfn;
790
791 printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
792 last_pfn, max_arch_pfn);
793 return last_pfn;
794}
795unsigned long __init e820_end_of_ram_pfn(void)
796{
797 return e820_end_pfn(MAX_ARCH_PFN);
798}
799
800unsigned long __init e820_end_of_low_ram_pfn(void)
801{
802 return e820_end_pfn(1UL << (32-PAGE_SHIFT));
803}
804
805static void early_panic(char *msg)
806{
807 early_printk(msg);
808 panic(msg);
809}
810
811static int userdef __initdata;
812
813/* "mem=nopentium" disables the 4MB page tables. */
814static int __init parse_memopt(char *p)
815{
816 u64 mem_size;
817
818 if (!p)
819 return -EINVAL;
820
821 if (!strcmp(p, "nopentium")) {
822#ifdef CONFIG_X86_32
823 setup_clear_cpu_cap(X86_FEATURE_PSE);
824 return 0;
825#else
826 printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
827 return -EINVAL;
828#endif
829 }
830
831 userdef = 1;
832 mem_size = memparse(p, &p);
833 /* don't remove all of memory when handling "mem={invalid}" param */
834 if (mem_size == 0)
835 return -EINVAL;
836 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
837
838 return 0;
839}
840early_param("mem", parse_memopt);
841
842static int __init parse_memmap_one(char *p)
843{
844 char *oldp;
845 u64 start_at, mem_size;
846
847 if (!p)
848 return -EINVAL;
849
850 if (!strncmp(p, "exactmap", 8)) {
851#ifdef CONFIG_CRASH_DUMP
852 /*
853 * If we are doing a crash dump, we still need to know
854 * the real mem size before original memory map is
855 * reset.
856 */
857 saved_max_pfn = e820_end_of_ram_pfn();
858#endif
859 e820.nr_map = 0;
860 userdef = 1;
861 return 0;
862 }
863
864 oldp = p;
865 mem_size = memparse(p, &p);
866 if (p == oldp)
867 return -EINVAL;
868
869 userdef = 1;
870 if (*p == '@') {
871 start_at = memparse(p+1, &p);
872 e820_add_region(start_at, mem_size, E820_RAM);
873 } else if (*p == '#') {
874 start_at = memparse(p+1, &p);
875 e820_add_region(start_at, mem_size, E820_ACPI);
876 } else if (*p == '$') {
877 start_at = memparse(p+1, &p);
878 e820_add_region(start_at, mem_size, E820_RESERVED);
879 } else if (*p == '!') {
880 start_at = memparse(p+1, &p);
881 e820_add_region(start_at, mem_size, E820_PRAM);
882 } else
883 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
884
885 return *p == '\0' ? 0 : -EINVAL;
886}
887static int __init parse_memmap_opt(char *str)
888{
889 while (str) {
890 char *k = strchr(str, ',');
891
892 if (k)
893 *k++ = 0;
894
895 parse_memmap_one(str);
896 str = k;
897 }
898
899 return 0;
900}
901early_param("memmap", parse_memmap_opt);
902
903void __init finish_e820_parsing(void)
904{
905 if (userdef) {
906 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map),
907 &e820.nr_map) < 0)
908 early_panic("Invalid user supplied memory map");
909
910 printk(KERN_INFO "e820: user-defined physical RAM map:\n");
911 e820_print_map("user");
912 }
913}
914
915static const char *e820_type_to_string(int e820_type)
916{
917 switch (e820_type) {
918 case E820_RESERVED_KERN:
919 case E820_RAM: return "System RAM";
920 case E820_ACPI: return "ACPI Tables";
921 case E820_NVS: return "ACPI Non-volatile Storage";
922 case E820_UNUSABLE: return "Unusable memory";
923 case E820_PRAM: return "Persistent Memory (legacy)";
924 case E820_PMEM: return "Persistent Memory";
925 default: return "reserved";
926 }
927}
928
929static unsigned long e820_type_to_iomem_type(int e820_type)
930{
931 switch (e820_type) {
932 case E820_RESERVED_KERN:
933 case E820_RAM:
934 return IORESOURCE_SYSTEM_RAM;
935 case E820_ACPI:
936 case E820_NVS:
937 case E820_UNUSABLE:
938 case E820_PRAM:
939 case E820_PMEM:
940 default:
941 return IORESOURCE_MEM;
942 }
943}
944
945static unsigned long e820_type_to_iores_desc(int e820_type)
946{
947 switch (e820_type) {
948 case E820_ACPI:
949 return IORES_DESC_ACPI_TABLES;
950 case E820_NVS:
951 return IORES_DESC_ACPI_NV_STORAGE;
952 case E820_PMEM:
953 return IORES_DESC_PERSISTENT_MEMORY;
954 case E820_PRAM:
955 return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
956 case E820_RESERVED_KERN:
957 case E820_RAM:
958 case E820_UNUSABLE:
959 default:
960 return IORES_DESC_NONE;
961 }
962}
963
964static bool do_mark_busy(u32 type, struct resource *res)
965{
966 /* this is the legacy bios/dos rom-shadow + mmio region */
967 if (res->start < (1ULL<<20))
968 return true;
969
970 /*
971 * Treat persistent memory like device memory, i.e. reserve it
972 * for exclusive use of a driver
973 */
974 switch (type) {
975 case E820_RESERVED:
976 case E820_PRAM:
977 case E820_PMEM:
978 return false;
979 default:
980 return true;
981 }
982}
983
984/*
985 * Mark e820 reserved areas as busy for the resource manager.
986 */
987static struct resource __initdata *e820_res;
988void __init e820_reserve_resources(void)
989{
990 int i;
991 struct resource *res;
992 u64 end;
993
994 res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
995 e820_res = res;
996 for (i = 0; i < e820.nr_map; i++) {
997 end = e820.map[i].addr + e820.map[i].size - 1;
998 if (end != (resource_size_t)end) {
999 res++;
1000 continue;
1001 }
1002 res->name = e820_type_to_string(e820.map[i].type);
1003 res->start = e820.map[i].addr;
1004 res->end = end;
1005
1006 res->flags = e820_type_to_iomem_type(e820.map[i].type);
1007 res->desc = e820_type_to_iores_desc(e820.map[i].type);
1008
1009 /*
1010 * don't register the region that could be conflicted with
1011 * pci device BAR resource and insert them later in
1012 * pcibios_resource_survey()
1013 */
1014 if (do_mark_busy(e820.map[i].type, res)) {
1015 res->flags |= IORESOURCE_BUSY;
1016 insert_resource(&iomem_resource, res);
1017 }
1018 res++;
1019 }
1020
1021 for (i = 0; i < e820_saved.nr_map; i++) {
1022 struct e820entry *entry = &e820_saved.map[i];
1023 firmware_map_add_early(entry->addr,
1024 entry->addr + entry->size,
1025 e820_type_to_string(entry->type));
1026 }
1027}
1028
1029/* How much should we pad RAM ending depending on where it is? */
1030static unsigned long ram_alignment(resource_size_t pos)
1031{
1032 unsigned long mb = pos >> 20;
1033
1034 /* To 64kB in the first megabyte */
1035 if (!mb)
1036 return 64*1024;
1037
1038 /* To 1MB in the first 16MB */
1039 if (mb < 16)
1040 return 1024*1024;
1041
1042 /* To 64MB for anything above that */
1043 return 64*1024*1024;
1044}
1045
1046#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1047
1048void __init e820_reserve_resources_late(void)
1049{
1050 int i;
1051 struct resource *res;
1052
1053 res = e820_res;
1054 for (i = 0; i < e820.nr_map; i++) {
1055 if (!res->parent && res->end)
1056 insert_resource_expand_to_fit(&iomem_resource, res);
1057 res++;
1058 }
1059
1060 /*
1061 * Try to bump up RAM regions to reasonable boundaries to
1062 * avoid stolen RAM:
1063 */
1064 for (i = 0; i < e820.nr_map; i++) {
1065 struct e820entry *entry = &e820.map[i];
1066 u64 start, end;
1067
1068 if (entry->type != E820_RAM)
1069 continue;
1070 start = entry->addr + entry->size;
1071 end = round_up(start, ram_alignment(start)) - 1;
1072 if (end > MAX_RESOURCE_SIZE)
1073 end = MAX_RESOURCE_SIZE;
1074 if (start >= end)
1075 continue;
1076 printk(KERN_DEBUG
1077 "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n",
1078 start, end);
1079 reserve_region_with_split(&iomem_resource, start, end,
1080 "RAM buffer");
1081 }
1082}
1083
1084char *__init default_machine_specific_memory_setup(void)
1085{
1086 char *who = "BIOS-e820";
1087 u32 new_nr;
1088 /*
1089 * Try to copy the BIOS-supplied E820-map.
1090 *
1091 * Otherwise fake a memory map; one section from 0k->640k,
1092 * the next section from 1mb->appropriate_mem_k
1093 */
1094 new_nr = boot_params.e820_entries;
1095 sanitize_e820_map(boot_params.e820_map,
1096 ARRAY_SIZE(boot_params.e820_map),
1097 &new_nr);
1098 boot_params.e820_entries = new_nr;
1099 if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1100 < 0) {
1101 u64 mem_size;
1102
1103 /* compare results from other methods and take the greater */
1104 if (boot_params.alt_mem_k
1105 < boot_params.screen_info.ext_mem_k) {
1106 mem_size = boot_params.screen_info.ext_mem_k;
1107 who = "BIOS-88";
1108 } else {
1109 mem_size = boot_params.alt_mem_k;
1110 who = "BIOS-e801";
1111 }
1112
1113 e820.nr_map = 0;
1114 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1115 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1116 }
1117
1118 /* In case someone cares... */
1119 return who;
1120}
1121
1122void __init setup_memory_map(void)
1123{
1124 char *who;
1125
1126 who = x86_init.resources.memory_setup();
1127 memcpy(&e820_saved, &e820, sizeof(struct e820map));
1128 printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n");
1129 e820_print_map(who);
1130}
1131
1132void __init memblock_x86_fill(void)
1133{
1134 int i;
1135 u64 end;
1136
1137 /*
1138 * EFI may have more than 128 entries
1139 * We are safe to enable resizing, beause memblock_x86_fill()
1140 * is rather later for x86
1141 */
1142 memblock_allow_resize();
1143
1144 for (i = 0; i < e820.nr_map; i++) {
1145 struct e820entry *ei = &e820.map[i];
1146
1147 end = ei->addr + ei->size;
1148 if (end != (resource_size_t)end)
1149 continue;
1150
1151 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1152 continue;
1153
1154 memblock_add(ei->addr, ei->size);
1155 }
1156
1157 /* throw away partial pages */
1158 memblock_trim_memory(PAGE_SIZE);
1159
1160 memblock_dump_all();
1161}
1162
1163void __init memblock_find_dma_reserve(void)
1164{
1165#ifdef CONFIG_X86_64
1166 u64 nr_pages = 0, nr_free_pages = 0;
1167 unsigned long start_pfn, end_pfn;
1168 phys_addr_t start, end;
1169 int i;
1170 u64 u;
1171
1172 /*
1173 * need to find out used area below MAX_DMA_PFN
1174 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1175 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1176 */
1177 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1178 start_pfn = min(start_pfn, MAX_DMA_PFN);
1179 end_pfn = min(end_pfn, MAX_DMA_PFN);
1180 nr_pages += end_pfn - start_pfn;
1181 }
1182
1183 for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1184 NULL) {
1185 start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1186 end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1187 if (start_pfn < end_pfn)
1188 nr_free_pages += end_pfn - start_pfn;
1189 }
1190
1191 set_dma_reserve(nr_pages - nr_free_pages);
1192#endif
1193}