hugetlb: introduce generic version of hugetlb_free_pgd_range
[linux/fpc-iii.git] / arch / x86 / kernel / e820.c
blobc88c23c658c1e99faad3daa236448bc4208901d7
1 /*
2 * Low level x86 E820 memory map handling functions.
4 * The firmware and bootloader passes us the "E820 table", which is the primary
5 * physical memory layout description available about x86 systems.
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.).
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>
19 #include <asm/e820/api.h>
20 #include <asm/setup.h>
23 * We organize the E820 table into three main data structures:
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:
30 * - inform the user about the firmware's notion of memory layout
31 * via /sys/firmware/memmap
33 * - the hibernation code uses it to generate a kernel-independent MD5
34 * fingerprint of the physical memory layout of a system.
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:
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.
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.
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.
58 static struct e820_table e820_table_init __initdata;
59 static struct e820_table e820_table_kexec_init __initdata;
60 static struct e820_table e820_table_firmware_init __initdata;
62 struct e820_table *e820_table __refdata = &e820_table_init;
63 struct e820_table *e820_table_kexec __refdata = &e820_table_kexec_init;
64 struct e820_table *e820_table_firmware __refdata = &e820_table_firmware_init;
66 /* For PCI or other memory-mapped resources */
67 unsigned long pci_mem_start = 0xaeedbabe;
68 #ifdef CONFIG_PCI
69 EXPORT_SYMBOL(pci_mem_start);
70 #endif
73 * This function checks if any part of the range <start,end> is mapped
74 * with type.
76 bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
78 int i;
80 for (i = 0; i < e820_table->nr_entries; i++) {
81 struct e820_entry *entry = &e820_table->entries[i];
83 if (type && entry->type != type)
84 continue;
85 if (entry->addr >= end || entry->addr + entry->size <= start)
86 continue;
87 return 1;
89 return 0;
91 EXPORT_SYMBOL_GPL(e820__mapped_any);
94 * This function checks if the entire <start,end> range is mapped with 'type'.
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.
99 static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
100 enum e820_type type)
102 int i;
104 for (i = 0; i < e820_table->nr_entries; i++) {
105 struct e820_entry *entry = &e820_table->entries[i];
107 if (type && entry->type != type)
108 continue;
110 /* Is the region (part) in overlap with the current region? */
111 if (entry->addr >= end || entry->addr + entry->size <= start)
112 continue;
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
118 if (entry->addr <= start)
119 start = entry->addr + entry->size;
122 * If 'start' is now at or beyond 'end', we're done, full
123 * coverage of the desired range exists:
125 if (start >= end)
126 return entry;
129 return NULL;
133 * This function checks if the entire range <start,end> is mapped with type.
135 bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
137 return __e820__mapped_all(start, end, type);
141 * This function returns the type associated with the range <start,end>.
143 int e820__get_entry_type(u64 start, u64 end)
145 struct e820_entry *entry = __e820__mapped_all(start, end, 0);
147 return entry ? entry->type : -EINVAL;
151 * Add a memory region to the kernel E820 map.
153 static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
155 int x = table->nr_entries;
157 if (x >= ARRAY_SIZE(table->entries)) {
158 pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
159 start, start + size - 1);
160 return;
163 table->entries[x].addr = start;
164 table->entries[x].size = size;
165 table->entries[x].type = type;
166 table->nr_entries++;
169 void __init e820__range_add(u64 start, u64 size, enum e820_type type)
171 __e820__range_add(e820_table, start, size, type);
174 static void __init e820_print_type(enum e820_type type)
176 switch (type) {
177 case E820_TYPE_RAM: /* Fall through: */
178 case E820_TYPE_RESERVED_KERN: pr_cont("usable"); break;
179 case E820_TYPE_RESERVED: pr_cont("reserved"); break;
180 case E820_TYPE_ACPI: pr_cont("ACPI data"); break;
181 case E820_TYPE_NVS: pr_cont("ACPI NVS"); break;
182 case E820_TYPE_UNUSABLE: pr_cont("unusable"); break;
183 case E820_TYPE_PMEM: /* Fall through: */
184 case E820_TYPE_PRAM: pr_cont("persistent (type %u)", type); break;
185 default: pr_cont("type %u", type); break;
189 void __init e820__print_table(char *who)
191 int i;
193 for (i = 0; i < e820_table->nr_entries; i++) {
194 pr_info("%s: [mem %#018Lx-%#018Lx] ",
195 who,
196 e820_table->entries[i].addr,
197 e820_table->entries[i].addr + e820_table->entries[i].size - 1);
199 e820_print_type(e820_table->entries[i].type);
200 pr_cont("\n");
205 * Sanitize an E820 map.
207 * Some E820 layouts include overlapping entries. The following
208 * replaces the original E820 map with a new one, removing overlaps,
209 * and resolving conflicting memory types in favor of highest
210 * numbered type.
212 * The input parameter 'entries' points to an array of 'struct
213 * e820_entry' which on entry has elements in the range [0, *nr_entries)
214 * valid, and which has space for up to max_nr_entries entries.
215 * On return, the resulting sanitized E820 map entries will be in
216 * overwritten in the same location, starting at 'entries'.
218 * The integer pointed to by nr_entries must be valid on entry (the
219 * current number of valid entries located at 'entries'). If the
220 * sanitizing succeeds the *nr_entries will be updated with the new
221 * number of valid entries (something no more than max_nr_entries).
223 * The return value from e820__update_table() is zero if it
224 * successfully 'sanitized' the map entries passed in, and is -1
225 * if it did nothing, which can happen if either of (1) it was
226 * only passed one map entry, or (2) any of the input map entries
227 * were invalid (start + size < start, meaning that the size was
228 * so big the described memory range wrapped around through zero.)
230 * Visually we're performing the following
231 * (1,2,3,4 = memory types)...
233 * Sample memory map (w/overlaps):
234 * ____22__________________
235 * ______________________4_
236 * ____1111________________
237 * _44_____________________
238 * 11111111________________
239 * ____________________33__
240 * ___________44___________
241 * __________33333_________
242 * ______________22________
243 * ___________________2222_
244 * _________111111111______
245 * _____________________11_
246 * _________________4______
248 * Sanitized equivalent (no overlap):
249 * 1_______________________
250 * _44_____________________
251 * ___1____________________
252 * ____22__________________
253 * ______11________________
254 * _________1______________
255 * __________3_____________
256 * ___________44___________
257 * _____________33_________
258 * _______________2________
259 * ________________1_______
260 * _________________4______
261 * ___________________2____
262 * ____________________33__
263 * ______________________4_
265 struct change_member {
266 /* Pointer to the original entry: */
267 struct e820_entry *entry;
268 /* Address for this change point: */
269 unsigned long long addr;
272 static struct change_member change_point_list[2*E820_MAX_ENTRIES] __initdata;
273 static struct change_member *change_point[2*E820_MAX_ENTRIES] __initdata;
274 static struct e820_entry *overlap_list[E820_MAX_ENTRIES] __initdata;
275 static struct e820_entry new_entries[E820_MAX_ENTRIES] __initdata;
277 static int __init cpcompare(const void *a, const void *b)
279 struct change_member * const *app = a, * const *bpp = b;
280 const struct change_member *ap = *app, *bp = *bpp;
283 * Inputs are pointers to two elements of change_point[]. If their
284 * addresses are not equal, their difference dominates. If the addresses
285 * are equal, then consider one that represents the end of its region
286 * to be greater than one that does not.
288 if (ap->addr != bp->addr)
289 return ap->addr > bp->addr ? 1 : -1;
291 return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
294 int __init e820__update_table(struct e820_table *table)
296 struct e820_entry *entries = table->entries;
297 u32 max_nr_entries = ARRAY_SIZE(table->entries);
298 enum e820_type current_type, last_type;
299 unsigned long long last_addr;
300 u32 new_nr_entries, overlap_entries;
301 u32 i, chg_idx, chg_nr;
303 /* If there's only one memory region, don't bother: */
304 if (table->nr_entries < 2)
305 return -1;
307 BUG_ON(table->nr_entries > max_nr_entries);
309 /* Bail out if we find any unreasonable addresses in the map: */
310 for (i = 0; i < table->nr_entries; i++) {
311 if (entries[i].addr + entries[i].size < entries[i].addr)
312 return -1;
315 /* Create pointers for initial change-point information (for sorting): */
316 for (i = 0; i < 2 * table->nr_entries; i++)
317 change_point[i] = &change_point_list[i];
320 * Record all known change-points (starting and ending addresses),
321 * omitting empty memory regions:
323 chg_idx = 0;
324 for (i = 0; i < table->nr_entries; i++) {
325 if (entries[i].size != 0) {
326 change_point[chg_idx]->addr = entries[i].addr;
327 change_point[chg_idx++]->entry = &entries[i];
328 change_point[chg_idx]->addr = entries[i].addr + entries[i].size;
329 change_point[chg_idx++]->entry = &entries[i];
332 chg_nr = chg_idx;
334 /* Sort change-point list by memory addresses (low -> high): */
335 sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
337 /* Create a new memory map, removing overlaps: */
338 overlap_entries = 0; /* Number of entries in the overlap table */
339 new_nr_entries = 0; /* Index for creating new map entries */
340 last_type = 0; /* Start with undefined memory type */
341 last_addr = 0; /* Start with 0 as last starting address */
343 /* Loop through change-points, determining effect on the new map: */
344 for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
345 /* Keep track of all overlapping entries */
346 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
347 /* Add map entry to overlap list (> 1 entry implies an overlap) */
348 overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
349 } else {
350 /* Remove entry from list (order independent, so swap with last): */
351 for (i = 0; i < overlap_entries; i++) {
352 if (overlap_list[i] == change_point[chg_idx]->entry)
353 overlap_list[i] = overlap_list[overlap_entries-1];
355 overlap_entries--;
358 * If there are overlapping entries, decide which
359 * "type" to use (larger value takes precedence --
360 * 1=usable, 2,3,4,4+=unusable)
362 current_type = 0;
363 for (i = 0; i < overlap_entries; i++) {
364 if (overlap_list[i]->type > current_type)
365 current_type = overlap_list[i]->type;
368 /* Continue building up new map based on this information: */
369 if (current_type != last_type || current_type == E820_TYPE_PRAM) {
370 if (last_type != 0) {
371 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
372 /* Move forward only if the new size was non-zero: */
373 if (new_entries[new_nr_entries].size != 0)
374 /* No more space left for new entries? */
375 if (++new_nr_entries >= max_nr_entries)
376 break;
378 if (current_type != 0) {
379 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
380 new_entries[new_nr_entries].type = current_type;
381 last_addr = change_point[chg_idx]->addr;
383 last_type = current_type;
387 /* Copy the new entries into the original location: */
388 memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
389 table->nr_entries = new_nr_entries;
391 return 0;
394 static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
396 struct boot_e820_entry *entry = entries;
398 while (nr_entries) {
399 u64 start = entry->addr;
400 u64 size = entry->size;
401 u64 end = start + size - 1;
402 u32 type = entry->type;
404 /* Ignore the entry on 64-bit overflow: */
405 if (start > end && likely(size))
406 return -1;
408 e820__range_add(start, size, type);
410 entry++;
411 nr_entries--;
413 return 0;
417 * Copy the BIOS E820 map into a safe place.
419 * Sanity-check it while we're at it..
421 * If we're lucky and live on a modern system, the setup code
422 * will have given us a memory map that we can use to properly
423 * set up memory. If we aren't, we'll fake a memory map.
425 static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
427 /* Only one memory region (or negative)? Ignore it */
428 if (nr_entries < 2)
429 return -1;
431 return __append_e820_table(entries, nr_entries);
434 static u64 __init
435 __e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
437 u64 end;
438 unsigned int i;
439 u64 real_updated_size = 0;
441 BUG_ON(old_type == new_type);
443 if (size > (ULLONG_MAX - start))
444 size = ULLONG_MAX - start;
446 end = start + size;
447 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
448 e820_print_type(old_type);
449 pr_cont(" ==> ");
450 e820_print_type(new_type);
451 pr_cont("\n");
453 for (i = 0; i < table->nr_entries; i++) {
454 struct e820_entry *entry = &table->entries[i];
455 u64 final_start, final_end;
456 u64 entry_end;
458 if (entry->type != old_type)
459 continue;
461 entry_end = entry->addr + entry->size;
463 /* Completely covered by new range? */
464 if (entry->addr >= start && entry_end <= end) {
465 entry->type = new_type;
466 real_updated_size += entry->size;
467 continue;
470 /* New range is completely covered? */
471 if (entry->addr < start && entry_end > end) {
472 __e820__range_add(table, start, size, new_type);
473 __e820__range_add(table, end, entry_end - end, entry->type);
474 entry->size = start - entry->addr;
475 real_updated_size += size;
476 continue;
479 /* Partially covered: */
480 final_start = max(start, entry->addr);
481 final_end = min(end, entry_end);
482 if (final_start >= final_end)
483 continue;
485 __e820__range_add(table, final_start, final_end - final_start, new_type);
487 real_updated_size += final_end - final_start;
490 * Left range could be head or tail, so need to update
491 * its size first:
493 entry->size -= final_end - final_start;
494 if (entry->addr < final_start)
495 continue;
497 entry->addr = final_end;
499 return real_updated_size;
502 u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
504 return __e820__range_update(e820_table, start, size, old_type, new_type);
507 static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
509 return __e820__range_update(e820_table_kexec, start, size, old_type, new_type);
512 /* Remove a range of memory from the E820 table: */
513 u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
515 int i;
516 u64 end;
517 u64 real_removed_size = 0;
519 if (size > (ULLONG_MAX - start))
520 size = ULLONG_MAX - start;
522 end = start + size;
523 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
524 if (check_type)
525 e820_print_type(old_type);
526 pr_cont("\n");
528 for (i = 0; i < e820_table->nr_entries; i++) {
529 struct e820_entry *entry = &e820_table->entries[i];
530 u64 final_start, final_end;
531 u64 entry_end;
533 if (check_type && entry->type != old_type)
534 continue;
536 entry_end = entry->addr + entry->size;
538 /* Completely covered? */
539 if (entry->addr >= start && entry_end <= end) {
540 real_removed_size += entry->size;
541 memset(entry, 0, sizeof(*entry));
542 continue;
545 /* Is the new range completely covered? */
546 if (entry->addr < start && entry_end > end) {
547 e820__range_add(end, entry_end - end, entry->type);
548 entry->size = start - entry->addr;
549 real_removed_size += size;
550 continue;
553 /* Partially covered: */
554 final_start = max(start, entry->addr);
555 final_end = min(end, entry_end);
556 if (final_start >= final_end)
557 continue;
559 real_removed_size += final_end - final_start;
562 * Left range could be head or tail, so need to update
563 * the size first:
565 entry->size -= final_end - final_start;
566 if (entry->addr < final_start)
567 continue;
569 entry->addr = final_end;
571 return real_removed_size;
574 void __init e820__update_table_print(void)
576 if (e820__update_table(e820_table))
577 return;
579 pr_info("modified physical RAM map:\n");
580 e820__print_table("modified");
583 static void __init e820__update_table_kexec(void)
585 e820__update_table(e820_table_kexec);
588 #define MAX_GAP_END 0x100000000ull
591 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
593 static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
595 unsigned long long last = MAX_GAP_END;
596 int i = e820_table->nr_entries;
597 int found = 0;
599 while (--i >= 0) {
600 unsigned long long start = e820_table->entries[i].addr;
601 unsigned long long end = start + e820_table->entries[i].size;
604 * Since "last" is at most 4GB, we know we'll
605 * fit in 32 bits if this condition is true:
607 if (last > end) {
608 unsigned long gap = last - end;
610 if (gap >= *gapsize) {
611 *gapsize = gap;
612 *gapstart = end;
613 found = 1;
616 if (start < last)
617 last = start;
619 return found;
623 * Search for the biggest gap in the low 32 bits of the E820
624 * memory space. We pass this space to the PCI subsystem, so
625 * that it can assign MMIO resources for hotplug or
626 * unconfigured devices in.
628 * Hopefully the BIOS let enough space left.
630 __init void e820__setup_pci_gap(void)
632 unsigned long gapstart, gapsize;
633 int found;
635 gapsize = 0x400000;
636 found = e820_search_gap(&gapstart, &gapsize);
638 if (!found) {
639 #ifdef CONFIG_X86_64
640 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
641 pr_err("Cannot find an available gap in the 32-bit address range\n");
642 pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
643 #else
644 gapstart = 0x10000000;
645 #endif
649 * e820__reserve_resources_late() protects stolen RAM already:
651 pci_mem_start = gapstart;
653 pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
654 gapstart, gapstart + gapsize - 1);
658 * Called late during init, in free_initmem().
660 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
662 * Copy them to a (usually much smaller) dynamically allocated area that is
663 * sized precisely after the number of e820 entries.
665 * This is done after we've performed all the fixes and tweaks to the tables.
666 * All functions which modify them are __init functions, which won't exist
667 * after free_initmem().
669 __init void e820__reallocate_tables(void)
671 struct e820_table *n;
672 int size;
674 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
675 n = kmalloc(size, GFP_KERNEL);
676 BUG_ON(!n);
677 memcpy(n, e820_table, size);
678 e820_table = n;
680 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
681 n = kmalloc(size, GFP_KERNEL);
682 BUG_ON(!n);
683 memcpy(n, e820_table_kexec, size);
684 e820_table_kexec = n;
686 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
687 n = kmalloc(size, GFP_KERNEL);
688 BUG_ON(!n);
689 memcpy(n, e820_table_firmware, size);
690 e820_table_firmware = n;
694 * Because of the small fixed size of struct boot_params, only the first
695 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
696 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
697 * struct setup_data, which is parsed here.
699 void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
701 int entries;
702 struct boot_e820_entry *extmap;
703 struct setup_data *sdata;
705 sdata = early_memremap(phys_addr, data_len);
706 entries = sdata->len / sizeof(*extmap);
707 extmap = (struct boot_e820_entry *)(sdata->data);
709 __append_e820_table(extmap, entries);
710 e820__update_table(e820_table);
712 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
713 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
715 early_memunmap(sdata, data_len);
716 pr_info("extended physical RAM map:\n");
717 e820__print_table("extended");
721 * Find the ranges of physical addresses that do not correspond to
722 * E820 RAM areas and register the corresponding pages as 'nosave' for
723 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
725 * This function requires the E820 map to be sorted and without any
726 * overlapping entries.
728 void __init e820__register_nosave_regions(unsigned long limit_pfn)
730 int i;
731 unsigned long pfn = 0;
733 for (i = 0; i < e820_table->nr_entries; i++) {
734 struct e820_entry *entry = &e820_table->entries[i];
736 if (pfn < PFN_UP(entry->addr))
737 register_nosave_region(pfn, PFN_UP(entry->addr));
739 pfn = PFN_DOWN(entry->addr + entry->size);
741 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
742 register_nosave_region(PFN_UP(entry->addr), pfn);
744 if (pfn >= limit_pfn)
745 break;
749 #ifdef CONFIG_ACPI
751 * Register ACPI NVS memory regions, so that we can save/restore them during
752 * hibernation and the subsequent resume:
754 static int __init e820__register_nvs_regions(void)
756 int i;
758 for (i = 0; i < e820_table->nr_entries; i++) {
759 struct e820_entry *entry = &e820_table->entries[i];
761 if (entry->type == E820_TYPE_NVS)
762 acpi_nvs_register(entry->addr, entry->size);
765 return 0;
767 core_initcall(e820__register_nvs_regions);
768 #endif
771 * Allocate the requested number of bytes with the requsted alignment
772 * and return (the physical address) to the caller. Also register this
773 * range in the 'kexec' E820 table as a reserved range.
775 * This allows kexec to fake a new mptable, as if it came from the real
776 * system.
778 u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
780 u64 addr;
782 addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
783 if (addr) {
784 e820__range_update_kexec(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
785 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
786 e820__update_table_kexec();
789 return addr;
792 #ifdef CONFIG_X86_32
793 # ifdef CONFIG_X86_PAE
794 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
795 # else
796 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
797 # endif
798 #else /* CONFIG_X86_32 */
799 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
800 #endif
803 * Find the highest page frame number we have available
805 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
807 int i;
808 unsigned long last_pfn = 0;
809 unsigned long max_arch_pfn = MAX_ARCH_PFN;
811 for (i = 0; i < e820_table->nr_entries; i++) {
812 struct e820_entry *entry = &e820_table->entries[i];
813 unsigned long start_pfn;
814 unsigned long end_pfn;
816 if (entry->type != type)
817 continue;
819 start_pfn = entry->addr >> PAGE_SHIFT;
820 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
822 if (start_pfn >= limit_pfn)
823 continue;
824 if (end_pfn > limit_pfn) {
825 last_pfn = limit_pfn;
826 break;
828 if (end_pfn > last_pfn)
829 last_pfn = end_pfn;
832 if (last_pfn > max_arch_pfn)
833 last_pfn = max_arch_pfn;
835 pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
836 last_pfn, max_arch_pfn);
837 return last_pfn;
840 unsigned long __init e820__end_of_ram_pfn(void)
842 return e820_end_pfn(MAX_ARCH_PFN, E820_TYPE_RAM);
845 unsigned long __init e820__end_of_low_ram_pfn(void)
847 return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
850 static void __init early_panic(char *msg)
852 early_printk(msg);
853 panic(msg);
856 static int userdef __initdata;
858 /* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
859 static int __init parse_memopt(char *p)
861 u64 mem_size;
863 if (!p)
864 return -EINVAL;
866 if (!strcmp(p, "nopentium")) {
867 #ifdef CONFIG_X86_32
868 setup_clear_cpu_cap(X86_FEATURE_PSE);
869 return 0;
870 #else
871 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
872 return -EINVAL;
873 #endif
876 userdef = 1;
877 mem_size = memparse(p, &p);
879 /* Don't remove all memory when getting "mem={invalid}" parameter: */
880 if (mem_size == 0)
881 return -EINVAL;
883 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
885 return 0;
887 early_param("mem", parse_memopt);
889 static int __init parse_memmap_one(char *p)
891 char *oldp;
892 u64 start_at, mem_size;
894 if (!p)
895 return -EINVAL;
897 if (!strncmp(p, "exactmap", 8)) {
898 #ifdef CONFIG_CRASH_DUMP
900 * If we are doing a crash dump, we still need to know
901 * the real memory size before the original memory map is
902 * reset.
904 saved_max_pfn = e820__end_of_ram_pfn();
905 #endif
906 e820_table->nr_entries = 0;
907 userdef = 1;
908 return 0;
911 oldp = p;
912 mem_size = memparse(p, &p);
913 if (p == oldp)
914 return -EINVAL;
916 userdef = 1;
917 if (*p == '@') {
918 start_at = memparse(p+1, &p);
919 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
920 } else if (*p == '#') {
921 start_at = memparse(p+1, &p);
922 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
923 } else if (*p == '$') {
924 start_at = memparse(p+1, &p);
925 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
926 } else if (*p == '!') {
927 start_at = memparse(p+1, &p);
928 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
929 } else if (*p == '%') {
930 enum e820_type from = 0, to = 0;
932 start_at = memparse(p + 1, &p);
933 if (*p == '-')
934 from = simple_strtoull(p + 1, &p, 0);
935 if (*p == '+')
936 to = simple_strtoull(p + 1, &p, 0);
937 if (*p != '\0')
938 return -EINVAL;
939 if (from && to)
940 e820__range_update(start_at, mem_size, from, to);
941 else if (to)
942 e820__range_add(start_at, mem_size, to);
943 else if (from)
944 e820__range_remove(start_at, mem_size, from, 1);
945 else
946 e820__range_remove(start_at, mem_size, 0, 0);
947 } else {
948 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
951 return *p == '\0' ? 0 : -EINVAL;
954 static int __init parse_memmap_opt(char *str)
956 while (str) {
957 char *k = strchr(str, ',');
959 if (k)
960 *k++ = 0;
962 parse_memmap_one(str);
963 str = k;
966 return 0;
968 early_param("memmap", parse_memmap_opt);
971 * Reserve all entries from the bootloader's extensible data nodes list,
972 * because if present we are going to use it later on to fetch e820
973 * entries from it:
975 void __init e820__reserve_setup_data(void)
977 struct setup_data *data;
978 u64 pa_data;
980 pa_data = boot_params.hdr.setup_data;
981 if (!pa_data)
982 return;
984 while (pa_data) {
985 data = early_memremap(pa_data, sizeof(*data));
986 e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
987 e820__range_update_kexec(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
988 pa_data = data->next;
989 early_memunmap(data, sizeof(*data));
992 e820__update_table(e820_table);
993 e820__update_table(e820_table_kexec);
995 pr_info("extended physical RAM map:\n");
996 e820__print_table("reserve setup_data");
1000 * Called after parse_early_param(), after early parameters (such as mem=)
1001 * have been processed, in which case we already have an E820 table filled in
1002 * via the parameter callback function(s), but it's not sorted and printed yet:
1004 void __init e820__finish_early_params(void)
1006 if (userdef) {
1007 if (e820__update_table(e820_table) < 0)
1008 early_panic("Invalid user supplied memory map");
1010 pr_info("user-defined physical RAM map:\n");
1011 e820__print_table("user");
1015 static const char *__init e820_type_to_string(struct e820_entry *entry)
1017 switch (entry->type) {
1018 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1019 case E820_TYPE_RAM: return "System RAM";
1020 case E820_TYPE_ACPI: return "ACPI Tables";
1021 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1022 case E820_TYPE_UNUSABLE: return "Unusable memory";
1023 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1024 case E820_TYPE_PMEM: return "Persistent Memory";
1025 case E820_TYPE_RESERVED: return "Reserved";
1026 default: return "Unknown E820 type";
1030 static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1032 switch (entry->type) {
1033 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1034 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1035 case E820_TYPE_ACPI: /* Fall-through: */
1036 case E820_TYPE_NVS: /* Fall-through: */
1037 case E820_TYPE_UNUSABLE: /* Fall-through: */
1038 case E820_TYPE_PRAM: /* Fall-through: */
1039 case E820_TYPE_PMEM: /* Fall-through: */
1040 case E820_TYPE_RESERVED: /* Fall-through: */
1041 default: return IORESOURCE_MEM;
1045 static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1047 switch (entry->type) {
1048 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1049 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1050 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1051 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1052 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1053 case E820_TYPE_RAM: /* Fall-through: */
1054 case E820_TYPE_UNUSABLE: /* Fall-through: */
1055 case E820_TYPE_RESERVED: /* Fall-through: */
1056 default: return IORES_DESC_NONE;
1060 static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1062 /* this is the legacy bios/dos rom-shadow + mmio region */
1063 if (res->start < (1ULL<<20))
1064 return true;
1067 * Treat persistent memory like device memory, i.e. reserve it
1068 * for exclusive use of a driver
1070 switch (type) {
1071 case E820_TYPE_RESERVED:
1072 case E820_TYPE_PRAM:
1073 case E820_TYPE_PMEM:
1074 return false;
1075 case E820_TYPE_RESERVED_KERN:
1076 case E820_TYPE_RAM:
1077 case E820_TYPE_ACPI:
1078 case E820_TYPE_NVS:
1079 case E820_TYPE_UNUSABLE:
1080 default:
1081 return true;
1086 * Mark E820 reserved areas as busy for the resource manager:
1089 static struct resource __initdata *e820_res;
1091 void __init e820__reserve_resources(void)
1093 int i;
1094 struct resource *res;
1095 u64 end;
1097 res = alloc_bootmem(sizeof(*res) * e820_table->nr_entries);
1098 e820_res = res;
1100 for (i = 0; i < e820_table->nr_entries; i++) {
1101 struct e820_entry *entry = e820_table->entries + i;
1103 end = entry->addr + entry->size - 1;
1104 if (end != (resource_size_t)end) {
1105 res++;
1106 continue;
1108 res->start = entry->addr;
1109 res->end = end;
1110 res->name = e820_type_to_string(entry);
1111 res->flags = e820_type_to_iomem_type(entry);
1112 res->desc = e820_type_to_iores_desc(entry);
1115 * Don't register the region that could be conflicted with
1116 * PCI device BAR resources and insert them later in
1117 * pcibios_resource_survey():
1119 if (do_mark_busy(entry->type, res)) {
1120 res->flags |= IORESOURCE_BUSY;
1121 insert_resource(&iomem_resource, res);
1123 res++;
1126 /* Expose the bootloader-provided memory layout to the sysfs. */
1127 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1128 struct e820_entry *entry = e820_table_firmware->entries + i;
1130 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1135 * How much should we pad the end of RAM, depending on where it is?
1137 static unsigned long __init ram_alignment(resource_size_t pos)
1139 unsigned long mb = pos >> 20;
1141 /* To 64kB in the first megabyte */
1142 if (!mb)
1143 return 64*1024;
1145 /* To 1MB in the first 16MB */
1146 if (mb < 16)
1147 return 1024*1024;
1149 /* To 64MB for anything above that */
1150 return 64*1024*1024;
1153 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1155 void __init e820__reserve_resources_late(void)
1157 int i;
1158 struct resource *res;
1160 res = e820_res;
1161 for (i = 0; i < e820_table->nr_entries; i++) {
1162 if (!res->parent && res->end)
1163 insert_resource_expand_to_fit(&iomem_resource, res);
1164 res++;
1168 * Try to bump up RAM regions to reasonable boundaries, to
1169 * avoid stolen RAM:
1171 for (i = 0; i < e820_table->nr_entries; i++) {
1172 struct e820_entry *entry = &e820_table->entries[i];
1173 u64 start, end;
1175 if (entry->type != E820_TYPE_RAM)
1176 continue;
1178 start = entry->addr + entry->size;
1179 end = round_up(start, ram_alignment(start)) - 1;
1180 if (end > MAX_RESOURCE_SIZE)
1181 end = MAX_RESOURCE_SIZE;
1182 if (start >= end)
1183 continue;
1185 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1186 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1191 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1193 char *__init e820__memory_setup_default(void)
1195 char *who = "BIOS-e820";
1198 * Try to copy the BIOS-supplied E820-map.
1200 * Otherwise fake a memory map; one section from 0k->640k,
1201 * the next section from 1mb->appropriate_mem_k
1203 if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1204 u64 mem_size;
1206 /* Compare results from other methods and take the one that gives more RAM: */
1207 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1208 mem_size = boot_params.screen_info.ext_mem_k;
1209 who = "BIOS-88";
1210 } else {
1211 mem_size = boot_params.alt_mem_k;
1212 who = "BIOS-e801";
1215 e820_table->nr_entries = 0;
1216 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1217 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1220 /* We just appended a lot of ranges, sanitize the table: */
1221 e820__update_table(e820_table);
1223 return who;
1227 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1228 * E820 map - with an optional platform quirk available for virtual platforms
1229 * to override this method of boot environment processing:
1231 void __init e820__memory_setup(void)
1233 char *who;
1235 /* This is a firmware interface ABI - make sure we don't break it: */
1236 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1238 who = x86_init.resources.memory_setup();
1240 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1241 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1243 pr_info("BIOS-provided physical RAM map:\n");
1244 e820__print_table(who);
1247 void __init e820__memblock_setup(void)
1249 int i;
1250 u64 end;
1251 u64 addr = 0;
1254 * The bootstrap memblock region count maximum is 128 entries
1255 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1256 * than that - so allow memblock resizing.
1258 * This is safe, because this call happens pretty late during x86 setup,
1259 * so we know about reserved memory regions already. (This is important
1260 * so that memblock resizing does no stomp over reserved areas.)
1262 memblock_allow_resize();
1264 for (i = 0; i < e820_table->nr_entries; i++) {
1265 struct e820_entry *entry = &e820_table->entries[i];
1267 end = entry->addr + entry->size;
1268 if (addr < entry->addr)
1269 memblock_reserve(addr, entry->addr - addr);
1270 addr = end;
1271 if (end != (resource_size_t)end)
1272 continue;
1275 * all !E820_TYPE_RAM ranges (including gap ranges) are put
1276 * into memblock.reserved to make sure that struct pages in
1277 * such regions are not left uninitialized after bootup.
1279 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1280 memblock_reserve(entry->addr, entry->size);
1281 else
1282 memblock_add(entry->addr, entry->size);
1285 /* Throw away partial pages: */
1286 memblock_trim_memory(PAGE_SIZE);
1288 memblock_dump_all();