1 // SPDX-License-Identifier: GPL-2.0-only
3 * Low level x86 E820 memory map handling functions.
5 * The firmware and bootloader passes us the "E820 table", which is the primary
6 * physical memory layout description available about x86 systems.
8 * The kernel takes the E820 memory layout and optionally modifies it with
9 * quirks and other tweaks, and feeds that into the generic Linux memory
10 * allocation code routines via a platform independent interface (memblock, etc.).
12 #include <linux/crash_dump.h>
13 #include <linux/memblock.h>
14 #include <linux/suspend.h>
15 #include <linux/acpi.h>
16 #include <linux/firmware-map.h>
17 #include <linux/sort.h>
18 #include <linux/memory_hotplug.h>
20 #include <asm/e820/api.h>
21 #include <asm/setup.h>
24 * We organize the E820 table into three main data structures:
26 * - 'e820_table_firmware': the original firmware version passed to us by the
27 * bootloader - not modified by the kernel. It is composed of two parts:
28 * the first 128 E820 memory entries in boot_params.e820_table and the remaining
29 * (if any) entries of the SETUP_E820_EXT nodes. We use this to:
31 * - inform the user about the firmware's notion of memory layout
32 * via /sys/firmware/memmap
34 * - the hibernation code uses it to generate a kernel-independent MD5
35 * fingerprint of the physical memory layout of a system.
37 * - 'e820_table_kexec': a slightly modified (by the kernel) firmware version
38 * passed to us by the bootloader - the major difference between
39 * e820_table_firmware[] and this one is that, the latter marks the setup_data
40 * list created by the EFI boot stub as reserved, so that kexec can reuse the
41 * setup_data information in the second kernel. Besides, e820_table_kexec[]
42 * might also be modified by the kexec itself to fake a mptable.
45 * - kexec, which is a bootloader in disguise, uses the original E820
46 * layout to pass to the kexec-ed kernel. This way the original kernel
47 * can have a restricted E820 map while the kexec()-ed kexec-kernel
48 * can have access to full memory - etc.
50 * - 'e820_table': this is the main E820 table that is massaged by the
51 * low level x86 platform code, or modified by boot parameters, before
52 * passed on to higher level MM layers.
54 * Once the E820 map has been converted to the standard Linux memory layout
55 * information its role stops - modifying it has no effect and does not get
56 * re-propagated. So itsmain role is a temporary bootstrap storage of firmware
57 * specific memory layout data during early bootup.
59 static struct e820_table e820_table_init __initdata;
60 static struct e820_table e820_table_kexec_init __initdata;
61 static struct e820_table e820_table_firmware_init __initdata;
63 struct e820_table *e820_table __refdata = &e820_table_init;
64 struct e820_table *e820_table_kexec __refdata = &e820_table_kexec_init;
65 struct e820_table *e820_table_firmware __refdata = &e820_table_firmware_init;
67 /* For PCI or other memory-mapped resources */
68 unsigned long pci_mem_start = 0xaeedbabe;
70 EXPORT_SYMBOL(pci_mem_start);
74 * This function checks if any part of the range <start,end> is mapped
77 static bool _e820__mapped_any(struct e820_table *table,
78 u64 start, u64 end, enum e820_type type)
82 for (i = 0; i < table->nr_entries; i++) {
83 struct e820_entry *entry = &table->entries[i];
85 if (type && entry->type != type)
87 if (entry->addr >= end || entry->addr + entry->size <= start)
94 bool e820__mapped_raw_any(u64 start, u64 end, enum e820_type type)
96 return _e820__mapped_any(e820_table_firmware, start, end, type);
98 EXPORT_SYMBOL_GPL(e820__mapped_raw_any);
100 bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
102 return _e820__mapped_any(e820_table, start, end, type);
104 EXPORT_SYMBOL_GPL(e820__mapped_any);
107 * This function checks if the entire <start,end> range is mapped with 'type'.
109 * Note: this function only works correctly once the E820 table is sorted and
110 * not-overlapping (at least for the range specified), which is the case normally.
112 static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
117 for (i = 0; i < e820_table->nr_entries; i++) {
118 struct e820_entry *entry = &e820_table->entries[i];
120 if (type && entry->type != type)
123 /* Is the region (part) in overlap with the current region? */
124 if (entry->addr >= end || entry->addr + entry->size <= start)
128 * If the region is at the beginning of <start,end> we move
129 * 'start' to the end of the region since it's ok until there
131 if (entry->addr <= start)
132 start = entry->addr + entry->size;
135 * If 'start' is now at or beyond 'end', we're done, full
136 * coverage of the desired range exists:
146 * This function checks if the entire range <start,end> is mapped with type.
148 bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
150 return __e820__mapped_all(start, end, type);
154 * This function returns the type associated with the range <start,end>.
156 int e820__get_entry_type(u64 start, u64 end)
158 struct e820_entry *entry = __e820__mapped_all(start, end, 0);
160 return entry ? entry->type : -EINVAL;
164 * Add a memory region to the kernel E820 map.
166 static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
168 int x = table->nr_entries;
170 if (x >= ARRAY_SIZE(table->entries)) {
171 pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
172 start, start + size - 1);
176 table->entries[x].addr = start;
177 table->entries[x].size = size;
178 table->entries[x].type = type;
182 void __init e820__range_add(u64 start, u64 size, enum e820_type type)
184 __e820__range_add(e820_table, start, size, type);
187 static void __init e820_print_type(enum e820_type type)
190 case E820_TYPE_RAM: /* Fall through: */
191 case E820_TYPE_RESERVED_KERN: pr_cont("usable"); break;
192 case E820_TYPE_RESERVED: pr_cont("reserved"); break;
193 case E820_TYPE_ACPI: pr_cont("ACPI data"); break;
194 case E820_TYPE_NVS: pr_cont("ACPI NVS"); break;
195 case E820_TYPE_UNUSABLE: pr_cont("unusable"); break;
196 case E820_TYPE_PMEM: /* Fall through: */
197 case E820_TYPE_PRAM: pr_cont("persistent (type %u)", type); break;
198 default: pr_cont("type %u", type); break;
202 void __init e820__print_table(char *who)
206 for (i = 0; i < e820_table->nr_entries; i++) {
207 pr_info("%s: [mem %#018Lx-%#018Lx] ",
209 e820_table->entries[i].addr,
210 e820_table->entries[i].addr + e820_table->entries[i].size - 1);
212 e820_print_type(e820_table->entries[i].type);
218 * Sanitize an E820 map.
220 * Some E820 layouts include overlapping entries. The following
221 * replaces the original E820 map with a new one, removing overlaps,
222 * and resolving conflicting memory types in favor of highest
225 * The input parameter 'entries' points to an array of 'struct
226 * e820_entry' which on entry has elements in the range [0, *nr_entries)
227 * valid, and which has space for up to max_nr_entries entries.
228 * On return, the resulting sanitized E820 map entries will be in
229 * overwritten in the same location, starting at 'entries'.
231 * The integer pointed to by nr_entries must be valid on entry (the
232 * current number of valid entries located at 'entries'). If the
233 * sanitizing succeeds the *nr_entries will be updated with the new
234 * number of valid entries (something no more than max_nr_entries).
236 * The return value from e820__update_table() is zero if it
237 * successfully 'sanitized' the map entries passed in, and is -1
238 * if it did nothing, which can happen if either of (1) it was
239 * only passed one map entry, or (2) any of the input map entries
240 * were invalid (start + size < start, meaning that the size was
241 * so big the described memory range wrapped around through zero.)
243 * Visually we're performing the following
244 * (1,2,3,4 = memory types)...
246 * Sample memory map (w/overlaps):
247 * ____22__________________
248 * ______________________4_
249 * ____1111________________
250 * _44_____________________
251 * 11111111________________
252 * ____________________33__
253 * ___________44___________
254 * __________33333_________
255 * ______________22________
256 * ___________________2222_
257 * _________111111111______
258 * _____________________11_
259 * _________________4______
261 * Sanitized equivalent (no overlap):
262 * 1_______________________
263 * _44_____________________
264 * ___1____________________
265 * ____22__________________
266 * ______11________________
267 * _________1______________
268 * __________3_____________
269 * ___________44___________
270 * _____________33_________
271 * _______________2________
272 * ________________1_______
273 * _________________4______
274 * ___________________2____
275 * ____________________33__
276 * ______________________4_
278 struct change_member {
279 /* Pointer to the original entry: */
280 struct e820_entry *entry;
281 /* Address for this change point: */
282 unsigned long long addr;
285 static struct change_member change_point_list[2*E820_MAX_ENTRIES] __initdata;
286 static struct change_member *change_point[2*E820_MAX_ENTRIES] __initdata;
287 static struct e820_entry *overlap_list[E820_MAX_ENTRIES] __initdata;
288 static struct e820_entry new_entries[E820_MAX_ENTRIES] __initdata;
290 static int __init cpcompare(const void *a, const void *b)
292 struct change_member * const *app = a, * const *bpp = b;
293 const struct change_member *ap = *app, *bp = *bpp;
296 * Inputs are pointers to two elements of change_point[]. If their
297 * addresses are not equal, their difference dominates. If the addresses
298 * are equal, then consider one that represents the end of its region
299 * to be greater than one that does not.
301 if (ap->addr != bp->addr)
302 return ap->addr > bp->addr ? 1 : -1;
304 return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
307 int __init e820__update_table(struct e820_table *table)
309 struct e820_entry *entries = table->entries;
310 u32 max_nr_entries = ARRAY_SIZE(table->entries);
311 enum e820_type current_type, last_type;
312 unsigned long long last_addr;
313 u32 new_nr_entries, overlap_entries;
314 u32 i, chg_idx, chg_nr;
316 /* If there's only one memory region, don't bother: */
317 if (table->nr_entries < 2)
320 BUG_ON(table->nr_entries > max_nr_entries);
322 /* Bail out if we find any unreasonable addresses in the map: */
323 for (i = 0; i < table->nr_entries; i++) {
324 if (entries[i].addr + entries[i].size < entries[i].addr)
328 /* Create pointers for initial change-point information (for sorting): */
329 for (i = 0; i < 2 * table->nr_entries; i++)
330 change_point[i] = &change_point_list[i];
333 * Record all known change-points (starting and ending addresses),
334 * omitting empty memory regions:
337 for (i = 0; i < table->nr_entries; i++) {
338 if (entries[i].size != 0) {
339 change_point[chg_idx]->addr = entries[i].addr;
340 change_point[chg_idx++]->entry = &entries[i];
341 change_point[chg_idx]->addr = entries[i].addr + entries[i].size;
342 change_point[chg_idx++]->entry = &entries[i];
347 /* Sort change-point list by memory addresses (low -> high): */
348 sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
350 /* Create a new memory map, removing overlaps: */
351 overlap_entries = 0; /* Number of entries in the overlap table */
352 new_nr_entries = 0; /* Index for creating new map entries */
353 last_type = 0; /* Start with undefined memory type */
354 last_addr = 0; /* Start with 0 as last starting address */
356 /* Loop through change-points, determining effect on the new map: */
357 for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
358 /* Keep track of all overlapping entries */
359 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
360 /* Add map entry to overlap list (> 1 entry implies an overlap) */
361 overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
363 /* Remove entry from list (order independent, so swap with last): */
364 for (i = 0; i < overlap_entries; i++) {
365 if (overlap_list[i] == change_point[chg_idx]->entry)
366 overlap_list[i] = overlap_list[overlap_entries-1];
371 * If there are overlapping entries, decide which
372 * "type" to use (larger value takes precedence --
373 * 1=usable, 2,3,4,4+=unusable)
376 for (i = 0; i < overlap_entries; i++) {
377 if (overlap_list[i]->type > current_type)
378 current_type = overlap_list[i]->type;
381 /* Continue building up new map based on this information: */
382 if (current_type != last_type || current_type == E820_TYPE_PRAM) {
383 if (last_type != 0) {
384 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
385 /* Move forward only if the new size was non-zero: */
386 if (new_entries[new_nr_entries].size != 0)
387 /* No more space left for new entries? */
388 if (++new_nr_entries >= max_nr_entries)
391 if (current_type != 0) {
392 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
393 new_entries[new_nr_entries].type = current_type;
394 last_addr = change_point[chg_idx]->addr;
396 last_type = current_type;
400 /* Copy the new entries into the original location: */
401 memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
402 table->nr_entries = new_nr_entries;
407 static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
409 struct boot_e820_entry *entry = entries;
412 u64 start = entry->addr;
413 u64 size = entry->size;
414 u64 end = start + size - 1;
415 u32 type = entry->type;
417 /* Ignore the entry on 64-bit overflow: */
418 if (start > end && likely(size))
421 e820__range_add(start, size, type);
430 * Copy the BIOS E820 map into a safe place.
432 * Sanity-check it while we're at it..
434 * If we're lucky and live on a modern system, the setup code
435 * will have given us a memory map that we can use to properly
436 * set up memory. If we aren't, we'll fake a memory map.
438 static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
440 /* Only one memory region (or negative)? Ignore it */
444 return __append_e820_table(entries, nr_entries);
448 __e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
452 u64 real_updated_size = 0;
454 BUG_ON(old_type == new_type);
456 if (size > (ULLONG_MAX - start))
457 size = ULLONG_MAX - start;
460 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
461 e820_print_type(old_type);
463 e820_print_type(new_type);
466 for (i = 0; i < table->nr_entries; i++) {
467 struct e820_entry *entry = &table->entries[i];
468 u64 final_start, final_end;
471 if (entry->type != old_type)
474 entry_end = entry->addr + entry->size;
476 /* Completely covered by new range? */
477 if (entry->addr >= start && entry_end <= end) {
478 entry->type = new_type;
479 real_updated_size += entry->size;
483 /* New range is completely covered? */
484 if (entry->addr < start && entry_end > end) {
485 __e820__range_add(table, start, size, new_type);
486 __e820__range_add(table, end, entry_end - end, entry->type);
487 entry->size = start - entry->addr;
488 real_updated_size += size;
492 /* Partially covered: */
493 final_start = max(start, entry->addr);
494 final_end = min(end, entry_end);
495 if (final_start >= final_end)
498 __e820__range_add(table, final_start, final_end - final_start, new_type);
500 real_updated_size += final_end - final_start;
503 * Left range could be head or tail, so need to update
506 entry->size -= final_end - final_start;
507 if (entry->addr < final_start)
510 entry->addr = final_end;
512 return real_updated_size;
515 u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
517 return __e820__range_update(e820_table, start, size, old_type, new_type);
520 static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
522 return __e820__range_update(e820_table_kexec, start, size, old_type, new_type);
525 /* Remove a range of memory from the E820 table: */
526 u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
530 u64 real_removed_size = 0;
532 if (size > (ULLONG_MAX - start))
533 size = ULLONG_MAX - start;
536 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
538 e820_print_type(old_type);
541 for (i = 0; i < e820_table->nr_entries; i++) {
542 struct e820_entry *entry = &e820_table->entries[i];
543 u64 final_start, final_end;
546 if (check_type && entry->type != old_type)
549 entry_end = entry->addr + entry->size;
551 /* Completely covered? */
552 if (entry->addr >= start && entry_end <= end) {
553 real_removed_size += entry->size;
554 memset(entry, 0, sizeof(*entry));
558 /* Is the new range completely covered? */
559 if (entry->addr < start && entry_end > end) {
560 e820__range_add(end, entry_end - end, entry->type);
561 entry->size = start - entry->addr;
562 real_removed_size += size;
566 /* Partially covered: */
567 final_start = max(start, entry->addr);
568 final_end = min(end, entry_end);
569 if (final_start >= final_end)
572 real_removed_size += final_end - final_start;
575 * Left range could be head or tail, so need to update
578 entry->size -= final_end - final_start;
579 if (entry->addr < final_start)
582 entry->addr = final_end;
584 return real_removed_size;
587 void __init e820__update_table_print(void)
589 if (e820__update_table(e820_table))
592 pr_info("modified physical RAM map:\n");
593 e820__print_table("modified");
596 static void __init e820__update_table_kexec(void)
598 e820__update_table(e820_table_kexec);
601 #define MAX_GAP_END 0x100000000ull
604 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
606 static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
608 unsigned long long last = MAX_GAP_END;
609 int i = e820_table->nr_entries;
613 unsigned long long start = e820_table->entries[i].addr;
614 unsigned long long end = start + e820_table->entries[i].size;
617 * Since "last" is at most 4GB, we know we'll
618 * fit in 32 bits if this condition is true:
621 unsigned long gap = last - end;
623 if (gap >= *gapsize) {
636 * Search for the biggest gap in the low 32 bits of the E820
637 * memory space. We pass this space to the PCI subsystem, so
638 * that it can assign MMIO resources for hotplug or
639 * unconfigured devices in.
641 * Hopefully the BIOS let enough space left.
643 __init void e820__setup_pci_gap(void)
645 unsigned long gapstart, gapsize;
649 found = e820_search_gap(&gapstart, &gapsize);
653 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
654 pr_err("Cannot find an available gap in the 32-bit address range\n");
655 pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
657 gapstart = 0x10000000;
662 * e820__reserve_resources_late() protects stolen RAM already:
664 pci_mem_start = gapstart;
666 pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
667 gapstart, gapstart + gapsize - 1);
671 * Called late during init, in free_initmem().
673 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
675 * Copy them to a (usually much smaller) dynamically allocated area that is
676 * sized precisely after the number of e820 entries.
678 * This is done after we've performed all the fixes and tweaks to the tables.
679 * All functions which modify them are __init functions, which won't exist
680 * after free_initmem().
682 __init void e820__reallocate_tables(void)
684 struct e820_table *n;
687 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
688 n = kmemdup(e820_table, size, GFP_KERNEL);
692 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
693 n = kmemdup(e820_table_kexec, size, GFP_KERNEL);
695 e820_table_kexec = n;
697 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
698 n = kmemdup(e820_table_firmware, size, GFP_KERNEL);
700 e820_table_firmware = n;
704 * Because of the small fixed size of struct boot_params, only the first
705 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
706 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
707 * struct setup_data, which is parsed here.
709 void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
712 struct boot_e820_entry *extmap;
713 struct setup_data *sdata;
715 sdata = early_memremap(phys_addr, data_len);
716 entries = sdata->len / sizeof(*extmap);
717 extmap = (struct boot_e820_entry *)(sdata->data);
719 __append_e820_table(extmap, entries);
720 e820__update_table(e820_table);
722 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
723 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
725 early_memunmap(sdata, data_len);
726 pr_info("extended physical RAM map:\n");
727 e820__print_table("extended");
731 * Find the ranges of physical addresses that do not correspond to
732 * E820 RAM areas and register the corresponding pages as 'nosave' for
733 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
735 * This function requires the E820 map to be sorted and without any
736 * overlapping entries.
738 void __init e820__register_nosave_regions(unsigned long limit_pfn)
741 unsigned long pfn = 0;
743 for (i = 0; i < e820_table->nr_entries; i++) {
744 struct e820_entry *entry = &e820_table->entries[i];
746 if (pfn < PFN_UP(entry->addr))
747 register_nosave_region(pfn, PFN_UP(entry->addr));
749 pfn = PFN_DOWN(entry->addr + entry->size);
751 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
752 register_nosave_region(PFN_UP(entry->addr), pfn);
754 if (pfn >= limit_pfn)
761 * Register ACPI NVS memory regions, so that we can save/restore them during
762 * hibernation and the subsequent resume:
764 static int __init e820__register_nvs_regions(void)
768 for (i = 0; i < e820_table->nr_entries; i++) {
769 struct e820_entry *entry = &e820_table->entries[i];
771 if (entry->type == E820_TYPE_NVS)
772 acpi_nvs_register(entry->addr, entry->size);
777 core_initcall(e820__register_nvs_regions);
781 * Allocate the requested number of bytes with the requsted alignment
782 * and return (the physical address) to the caller. Also register this
783 * range in the 'kexec' E820 table as a reserved range.
785 * This allows kexec to fake a new mptable, as if it came from the real
788 u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
792 addr = memblock_phys_alloc(size, align);
794 e820__range_update_kexec(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
795 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
796 e820__update_table_kexec();
803 # ifdef CONFIG_X86_PAE
804 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
806 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
808 #else /* CONFIG_X86_32 */
809 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
813 * Find the highest page frame number we have available
815 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
818 unsigned long last_pfn = 0;
819 unsigned long max_arch_pfn = MAX_ARCH_PFN;
821 for (i = 0; i < e820_table->nr_entries; i++) {
822 struct e820_entry *entry = &e820_table->entries[i];
823 unsigned long start_pfn;
824 unsigned long end_pfn;
826 if (entry->type != type)
829 start_pfn = entry->addr >> PAGE_SHIFT;
830 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
832 if (start_pfn >= limit_pfn)
834 if (end_pfn > limit_pfn) {
835 last_pfn = limit_pfn;
838 if (end_pfn > last_pfn)
842 if (last_pfn > max_arch_pfn)
843 last_pfn = max_arch_pfn;
845 pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
846 last_pfn, max_arch_pfn);
850 unsigned long __init e820__end_of_ram_pfn(void)
852 return e820_end_pfn(MAX_ARCH_PFN, E820_TYPE_RAM);
855 unsigned long __init e820__end_of_low_ram_pfn(void)
857 return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
860 static void __init early_panic(char *msg)
866 static int userdef __initdata;
868 /* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
869 static int __init parse_memopt(char *p)
876 if (!strcmp(p, "nopentium")) {
878 setup_clear_cpu_cap(X86_FEATURE_PSE);
881 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
887 mem_size = memparse(p, &p);
889 /* Don't remove all memory when getting "mem={invalid}" parameter: */
893 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
895 #ifdef CONFIG_MEMORY_HOTPLUG
896 max_mem_size = mem_size;
901 early_param("mem", parse_memopt);
903 static int __init parse_memmap_one(char *p)
906 u64 start_at, mem_size;
911 if (!strncmp(p, "exactmap", 8)) {
912 #ifdef CONFIG_CRASH_DUMP
914 * If we are doing a crash dump, we still need to know
915 * the real memory size before the original memory map is
918 saved_max_pfn = e820__end_of_ram_pfn();
920 e820_table->nr_entries = 0;
926 mem_size = memparse(p, &p);
932 start_at = memparse(p+1, &p);
933 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
934 } else if (*p == '#') {
935 start_at = memparse(p+1, &p);
936 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
937 } else if (*p == '$') {
938 start_at = memparse(p+1, &p);
939 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
940 } else if (*p == '!') {
941 start_at = memparse(p+1, &p);
942 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
943 } else if (*p == '%') {
944 enum e820_type from = 0, to = 0;
946 start_at = memparse(p + 1, &p);
948 from = simple_strtoull(p + 1, &p, 0);
950 to = simple_strtoull(p + 1, &p, 0);
954 e820__range_update(start_at, mem_size, from, to);
956 e820__range_add(start_at, mem_size, to);
958 e820__range_remove(start_at, mem_size, from, 1);
960 e820__range_remove(start_at, mem_size, 0, 0);
962 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
965 return *p == '\0' ? 0 : -EINVAL;
968 static int __init parse_memmap_opt(char *str)
971 char *k = strchr(str, ',');
976 parse_memmap_one(str);
982 early_param("memmap", parse_memmap_opt);
985 * Reserve all entries from the bootloader's extensible data nodes list,
986 * because if present we are going to use it later on to fetch e820
989 void __init e820__reserve_setup_data(void)
991 struct setup_data *data;
994 pa_data = boot_params.hdr.setup_data;
999 data = early_memremap(pa_data, sizeof(*data));
1000 e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1001 e820__range_update_kexec(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1002 pa_data = data->next;
1003 early_memunmap(data, sizeof(*data));
1006 e820__update_table(e820_table);
1007 e820__update_table(e820_table_kexec);
1009 pr_info("extended physical RAM map:\n");
1010 e820__print_table("reserve setup_data");
1014 * Called after parse_early_param(), after early parameters (such as mem=)
1015 * have been processed, in which case we already have an E820 table filled in
1016 * via the parameter callback function(s), but it's not sorted and printed yet:
1018 void __init e820__finish_early_params(void)
1021 if (e820__update_table(e820_table) < 0)
1022 early_panic("Invalid user supplied memory map");
1024 pr_info("user-defined physical RAM map:\n");
1025 e820__print_table("user");
1029 static const char *__init e820_type_to_string(struct e820_entry *entry)
1031 switch (entry->type) {
1032 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1033 case E820_TYPE_RAM: return "System RAM";
1034 case E820_TYPE_ACPI: return "ACPI Tables";
1035 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1036 case E820_TYPE_UNUSABLE: return "Unusable memory";
1037 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1038 case E820_TYPE_PMEM: return "Persistent Memory";
1039 case E820_TYPE_RESERVED: return "Reserved";
1040 default: return "Unknown E820 type";
1044 static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1046 switch (entry->type) {
1047 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1048 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1049 case E820_TYPE_ACPI: /* Fall-through: */
1050 case E820_TYPE_NVS: /* Fall-through: */
1051 case E820_TYPE_UNUSABLE: /* Fall-through: */
1052 case E820_TYPE_PRAM: /* Fall-through: */
1053 case E820_TYPE_PMEM: /* Fall-through: */
1054 case E820_TYPE_RESERVED: /* Fall-through: */
1055 default: return IORESOURCE_MEM;
1059 static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1061 switch (entry->type) {
1062 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1063 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1064 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1065 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1066 case E820_TYPE_RESERVED: return IORES_DESC_RESERVED;
1067 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1068 case E820_TYPE_RAM: /* Fall-through: */
1069 case E820_TYPE_UNUSABLE: /* Fall-through: */
1070 default: return IORES_DESC_NONE;
1074 static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1076 /* this is the legacy bios/dos rom-shadow + mmio region */
1077 if (res->start < (1ULL<<20))
1081 * Treat persistent memory like device memory, i.e. reserve it
1082 * for exclusive use of a driver
1085 case E820_TYPE_RESERVED:
1086 case E820_TYPE_PRAM:
1087 case E820_TYPE_PMEM:
1089 case E820_TYPE_RESERVED_KERN:
1091 case E820_TYPE_ACPI:
1093 case E820_TYPE_UNUSABLE:
1100 * Mark E820 reserved areas as busy for the resource manager:
1103 static struct resource __initdata *e820_res;
1105 void __init e820__reserve_resources(void)
1108 struct resource *res;
1111 res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1114 panic("%s: Failed to allocate %zu bytes\n", __func__,
1115 sizeof(*res) * e820_table->nr_entries);
1118 for (i = 0; i < e820_table->nr_entries; i++) {
1119 struct e820_entry *entry = e820_table->entries + i;
1121 end = entry->addr + entry->size - 1;
1122 if (end != (resource_size_t)end) {
1126 res->start = entry->addr;
1128 res->name = e820_type_to_string(entry);
1129 res->flags = e820_type_to_iomem_type(entry);
1130 res->desc = e820_type_to_iores_desc(entry);
1133 * Don't register the region that could be conflicted with
1134 * PCI device BAR resources and insert them later in
1135 * pcibios_resource_survey():
1137 if (do_mark_busy(entry->type, res)) {
1138 res->flags |= IORESOURCE_BUSY;
1139 insert_resource(&iomem_resource, res);
1144 /* Expose the bootloader-provided memory layout to the sysfs. */
1145 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1146 struct e820_entry *entry = e820_table_firmware->entries + i;
1148 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1153 * How much should we pad the end of RAM, depending on where it is?
1155 static unsigned long __init ram_alignment(resource_size_t pos)
1157 unsigned long mb = pos >> 20;
1159 /* To 64kB in the first megabyte */
1163 /* To 1MB in the first 16MB */
1167 /* To 64MB for anything above that */
1168 return 64*1024*1024;
1171 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1173 void __init e820__reserve_resources_late(void)
1176 struct resource *res;
1179 for (i = 0; i < e820_table->nr_entries; i++) {
1180 if (!res->parent && res->end)
1181 insert_resource_expand_to_fit(&iomem_resource, res);
1186 * Try to bump up RAM regions to reasonable boundaries, to
1189 for (i = 0; i < e820_table->nr_entries; i++) {
1190 struct e820_entry *entry = &e820_table->entries[i];
1193 if (entry->type != E820_TYPE_RAM)
1196 start = entry->addr + entry->size;
1197 end = round_up(start, ram_alignment(start)) - 1;
1198 if (end > MAX_RESOURCE_SIZE)
1199 end = MAX_RESOURCE_SIZE;
1203 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1204 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1209 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1211 char *__init e820__memory_setup_default(void)
1213 char *who = "BIOS-e820";
1216 * Try to copy the BIOS-supplied E820-map.
1218 * Otherwise fake a memory map; one section from 0k->640k,
1219 * the next section from 1mb->appropriate_mem_k
1221 if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1224 /* Compare results from other methods and take the one that gives more RAM: */
1225 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1226 mem_size = boot_params.screen_info.ext_mem_k;
1229 mem_size = boot_params.alt_mem_k;
1233 e820_table->nr_entries = 0;
1234 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1235 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1238 /* We just appended a lot of ranges, sanitize the table: */
1239 e820__update_table(e820_table);
1245 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1246 * E820 map - with an optional platform quirk available for virtual platforms
1247 * to override this method of boot environment processing:
1249 void __init e820__memory_setup(void)
1253 /* This is a firmware interface ABI - make sure we don't break it: */
1254 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1256 who = x86_init.resources.memory_setup();
1258 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1259 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1261 pr_info("BIOS-provided physical RAM map:\n");
1262 e820__print_table(who);
1265 void __init e820__memblock_setup(void)
1271 * The bootstrap memblock region count maximum is 128 entries
1272 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1273 * than that - so allow memblock resizing.
1275 * This is safe, because this call happens pretty late during x86 setup,
1276 * so we know about reserved memory regions already. (This is important
1277 * so that memblock resizing does no stomp over reserved areas.)
1279 memblock_allow_resize();
1281 for (i = 0; i < e820_table->nr_entries; i++) {
1282 struct e820_entry *entry = &e820_table->entries[i];
1284 end = entry->addr + entry->size;
1285 if (end != (resource_size_t)end)
1288 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1291 memblock_add(entry->addr, entry->size);
1294 /* Throw away partial pages: */
1295 memblock_trim_memory(PAGE_SIZE);
1297 memblock_dump_all();