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_SOFT_RESERVED: pr_cont("soft reserved"); break;
194 case E820_TYPE_ACPI: pr_cont("ACPI data"); break;
195 case E820_TYPE_NVS: pr_cont("ACPI NVS"); break;
196 case E820_TYPE_UNUSABLE: pr_cont("unusable"); break;
197 case E820_TYPE_PMEM: /* Fall through: */
198 case E820_TYPE_PRAM: pr_cont("persistent (type %u)", type); break;
199 default: pr_cont("type %u", type); break;
203 void __init e820__print_table(char *who)
207 for (i = 0; i < e820_table->nr_entries; i++) {
208 pr_info("%s: [mem %#018Lx-%#018Lx] ",
210 e820_table->entries[i].addr,
211 e820_table->entries[i].addr + e820_table->entries[i].size - 1);
213 e820_print_type(e820_table->entries[i].type);
219 * Sanitize an E820 map.
221 * Some E820 layouts include overlapping entries. The following
222 * replaces the original E820 map with a new one, removing overlaps,
223 * and resolving conflicting memory types in favor of highest
226 * The input parameter 'entries' points to an array of 'struct
227 * e820_entry' which on entry has elements in the range [0, *nr_entries)
228 * valid, and which has space for up to max_nr_entries entries.
229 * On return, the resulting sanitized E820 map entries will be in
230 * overwritten in the same location, starting at 'entries'.
232 * The integer pointed to by nr_entries must be valid on entry (the
233 * current number of valid entries located at 'entries'). If the
234 * sanitizing succeeds the *nr_entries will be updated with the new
235 * number of valid entries (something no more than max_nr_entries).
237 * The return value from e820__update_table() is zero if it
238 * successfully 'sanitized' the map entries passed in, and is -1
239 * if it did nothing, which can happen if either of (1) it was
240 * only passed one map entry, or (2) any of the input map entries
241 * were invalid (start + size < start, meaning that the size was
242 * so big the described memory range wrapped around through zero.)
244 * Visually we're performing the following
245 * (1,2,3,4 = memory types)...
247 * Sample memory map (w/overlaps):
248 * ____22__________________
249 * ______________________4_
250 * ____1111________________
251 * _44_____________________
252 * 11111111________________
253 * ____________________33__
254 * ___________44___________
255 * __________33333_________
256 * ______________22________
257 * ___________________2222_
258 * _________111111111______
259 * _____________________11_
260 * _________________4______
262 * Sanitized equivalent (no overlap):
263 * 1_______________________
264 * _44_____________________
265 * ___1____________________
266 * ____22__________________
267 * ______11________________
268 * _________1______________
269 * __________3_____________
270 * ___________44___________
271 * _____________33_________
272 * _______________2________
273 * ________________1_______
274 * _________________4______
275 * ___________________2____
276 * ____________________33__
277 * ______________________4_
279 struct change_member {
280 /* Pointer to the original entry: */
281 struct e820_entry *entry;
282 /* Address for this change point: */
283 unsigned long long addr;
286 static struct change_member change_point_list[2*E820_MAX_ENTRIES] __initdata;
287 static struct change_member *change_point[2*E820_MAX_ENTRIES] __initdata;
288 static struct e820_entry *overlap_list[E820_MAX_ENTRIES] __initdata;
289 static struct e820_entry new_entries[E820_MAX_ENTRIES] __initdata;
291 static int __init cpcompare(const void *a, const void *b)
293 struct change_member * const *app = a, * const *bpp = b;
294 const struct change_member *ap = *app, *bp = *bpp;
297 * Inputs are pointers to two elements of change_point[]. If their
298 * addresses are not equal, their difference dominates. If the addresses
299 * are equal, then consider one that represents the end of its region
300 * to be greater than one that does not.
302 if (ap->addr != bp->addr)
303 return ap->addr > bp->addr ? 1 : -1;
305 return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
308 int __init e820__update_table(struct e820_table *table)
310 struct e820_entry *entries = table->entries;
311 u32 max_nr_entries = ARRAY_SIZE(table->entries);
312 enum e820_type current_type, last_type;
313 unsigned long long last_addr;
314 u32 new_nr_entries, overlap_entries;
315 u32 i, chg_idx, chg_nr;
317 /* If there's only one memory region, don't bother: */
318 if (table->nr_entries < 2)
321 BUG_ON(table->nr_entries > max_nr_entries);
323 /* Bail out if we find any unreasonable addresses in the map: */
324 for (i = 0; i < table->nr_entries; i++) {
325 if (entries[i].addr + entries[i].size < entries[i].addr)
329 /* Create pointers for initial change-point information (for sorting): */
330 for (i = 0; i < 2 * table->nr_entries; i++)
331 change_point[i] = &change_point_list[i];
334 * Record all known change-points (starting and ending addresses),
335 * omitting empty memory regions:
338 for (i = 0; i < table->nr_entries; i++) {
339 if (entries[i].size != 0) {
340 change_point[chg_idx]->addr = entries[i].addr;
341 change_point[chg_idx++]->entry = &entries[i];
342 change_point[chg_idx]->addr = entries[i].addr + entries[i].size;
343 change_point[chg_idx++]->entry = &entries[i];
348 /* Sort change-point list by memory addresses (low -> high): */
349 sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
351 /* Create a new memory map, removing overlaps: */
352 overlap_entries = 0; /* Number of entries in the overlap table */
353 new_nr_entries = 0; /* Index for creating new map entries */
354 last_type = 0; /* Start with undefined memory type */
355 last_addr = 0; /* Start with 0 as last starting address */
357 /* Loop through change-points, determining effect on the new map: */
358 for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
359 /* Keep track of all overlapping entries */
360 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
361 /* Add map entry to overlap list (> 1 entry implies an overlap) */
362 overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
364 /* Remove entry from list (order independent, so swap with last): */
365 for (i = 0; i < overlap_entries; i++) {
366 if (overlap_list[i] == change_point[chg_idx]->entry)
367 overlap_list[i] = overlap_list[overlap_entries-1];
372 * If there are overlapping entries, decide which
373 * "type" to use (larger value takes precedence --
374 * 1=usable, 2,3,4,4+=unusable)
377 for (i = 0; i < overlap_entries; i++) {
378 if (overlap_list[i]->type > current_type)
379 current_type = overlap_list[i]->type;
382 /* Continue building up new map based on this information: */
383 if (current_type != last_type || current_type == E820_TYPE_PRAM) {
384 if (last_type != 0) {
385 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
386 /* Move forward only if the new size was non-zero: */
387 if (new_entries[new_nr_entries].size != 0)
388 /* No more space left for new entries? */
389 if (++new_nr_entries >= max_nr_entries)
392 if (current_type != 0) {
393 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
394 new_entries[new_nr_entries].type = current_type;
395 last_addr = change_point[chg_idx]->addr;
397 last_type = current_type;
401 /* Copy the new entries into the original location: */
402 memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
403 table->nr_entries = new_nr_entries;
408 static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
410 struct boot_e820_entry *entry = entries;
413 u64 start = entry->addr;
414 u64 size = entry->size;
415 u64 end = start + size - 1;
416 u32 type = entry->type;
418 /* Ignore the entry on 64-bit overflow: */
419 if (start > end && likely(size))
422 e820__range_add(start, size, type);
431 * Copy the BIOS E820 map into a safe place.
433 * Sanity-check it while we're at it..
435 * If we're lucky and live on a modern system, the setup code
436 * will have given us a memory map that we can use to properly
437 * set up memory. If we aren't, we'll fake a memory map.
439 static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
441 /* Only one memory region (or negative)? Ignore it */
445 return __append_e820_table(entries, nr_entries);
449 __e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
453 u64 real_updated_size = 0;
455 BUG_ON(old_type == new_type);
457 if (size > (ULLONG_MAX - start))
458 size = ULLONG_MAX - start;
461 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
462 e820_print_type(old_type);
464 e820_print_type(new_type);
467 for (i = 0; i < table->nr_entries; i++) {
468 struct e820_entry *entry = &table->entries[i];
469 u64 final_start, final_end;
472 if (entry->type != old_type)
475 entry_end = entry->addr + entry->size;
477 /* Completely covered by new range? */
478 if (entry->addr >= start && entry_end <= end) {
479 entry->type = new_type;
480 real_updated_size += entry->size;
484 /* New range is completely covered? */
485 if (entry->addr < start && entry_end > end) {
486 __e820__range_add(table, start, size, new_type);
487 __e820__range_add(table, end, entry_end - end, entry->type);
488 entry->size = start - entry->addr;
489 real_updated_size += size;
493 /* Partially covered: */
494 final_start = max(start, entry->addr);
495 final_end = min(end, entry_end);
496 if (final_start >= final_end)
499 __e820__range_add(table, final_start, final_end - final_start, new_type);
501 real_updated_size += final_end - final_start;
504 * Left range could be head or tail, so need to update
507 entry->size -= final_end - final_start;
508 if (entry->addr < final_start)
511 entry->addr = final_end;
513 return real_updated_size;
516 u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
518 return __e820__range_update(e820_table, start, size, old_type, new_type);
521 static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
523 return __e820__range_update(e820_table_kexec, start, size, old_type, new_type);
526 /* Remove a range of memory from the E820 table: */
527 u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
531 u64 real_removed_size = 0;
533 if (size > (ULLONG_MAX - start))
534 size = ULLONG_MAX - start;
537 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
539 e820_print_type(old_type);
542 for (i = 0; i < e820_table->nr_entries; i++) {
543 struct e820_entry *entry = &e820_table->entries[i];
544 u64 final_start, final_end;
547 if (check_type && entry->type != old_type)
550 entry_end = entry->addr + entry->size;
552 /* Completely covered? */
553 if (entry->addr >= start && entry_end <= end) {
554 real_removed_size += entry->size;
555 memset(entry, 0, sizeof(*entry));
559 /* Is the new range completely covered? */
560 if (entry->addr < start && entry_end > end) {
561 e820__range_add(end, entry_end - end, entry->type);
562 entry->size = start - entry->addr;
563 real_removed_size += size;
567 /* Partially covered: */
568 final_start = max(start, entry->addr);
569 final_end = min(end, entry_end);
570 if (final_start >= final_end)
573 real_removed_size += final_end - final_start;
576 * Left range could be head or tail, so need to update
579 entry->size -= final_end - final_start;
580 if (entry->addr < final_start)
583 entry->addr = final_end;
585 return real_removed_size;
588 void __init e820__update_table_print(void)
590 if (e820__update_table(e820_table))
593 pr_info("modified physical RAM map:\n");
594 e820__print_table("modified");
597 static void __init e820__update_table_kexec(void)
599 e820__update_table(e820_table_kexec);
602 #define MAX_GAP_END 0x100000000ull
605 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
607 static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
609 unsigned long long last = MAX_GAP_END;
610 int i = e820_table->nr_entries;
614 unsigned long long start = e820_table->entries[i].addr;
615 unsigned long long end = start + e820_table->entries[i].size;
618 * Since "last" is at most 4GB, we know we'll
619 * fit in 32 bits if this condition is true:
622 unsigned long gap = last - end;
624 if (gap >= *gapsize) {
637 * Search for the biggest gap in the low 32 bits of the E820
638 * memory space. We pass this space to the PCI subsystem, so
639 * that it can assign MMIO resources for hotplug or
640 * unconfigured devices in.
642 * Hopefully the BIOS let enough space left.
644 __init void e820__setup_pci_gap(void)
646 unsigned long gapstart, gapsize;
650 found = e820_search_gap(&gapstart, &gapsize);
654 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
655 pr_err("Cannot find an available gap in the 32-bit address range\n");
656 pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
658 gapstart = 0x10000000;
663 * e820__reserve_resources_late() protects stolen RAM already:
665 pci_mem_start = gapstart;
667 pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
668 gapstart, gapstart + gapsize - 1);
672 * Called late during init, in free_initmem().
674 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
676 * Copy them to a (usually much smaller) dynamically allocated area that is
677 * sized precisely after the number of e820 entries.
679 * This is done after we've performed all the fixes and tweaks to the tables.
680 * All functions which modify them are __init functions, which won't exist
681 * after free_initmem().
683 __init void e820__reallocate_tables(void)
685 struct e820_table *n;
688 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
689 n = kmemdup(e820_table, size, GFP_KERNEL);
693 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
694 n = kmemdup(e820_table_kexec, size, GFP_KERNEL);
696 e820_table_kexec = n;
698 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
699 n = kmemdup(e820_table_firmware, size, GFP_KERNEL);
701 e820_table_firmware = n;
705 * Because of the small fixed size of struct boot_params, only the first
706 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
707 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
708 * struct setup_data, which is parsed here.
710 void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
713 struct boot_e820_entry *extmap;
714 struct setup_data *sdata;
716 sdata = early_memremap(phys_addr, data_len);
717 entries = sdata->len / sizeof(*extmap);
718 extmap = (struct boot_e820_entry *)(sdata->data);
720 __append_e820_table(extmap, entries);
721 e820__update_table(e820_table);
723 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
724 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
726 early_memunmap(sdata, data_len);
727 pr_info("extended physical RAM map:\n");
728 e820__print_table("extended");
732 * Find the ranges of physical addresses that do not correspond to
733 * E820 RAM areas and register the corresponding pages as 'nosave' for
734 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
736 * This function requires the E820 map to be sorted and without any
737 * overlapping entries.
739 void __init e820__register_nosave_regions(unsigned long limit_pfn)
742 unsigned long pfn = 0;
744 for (i = 0; i < e820_table->nr_entries; i++) {
745 struct e820_entry *entry = &e820_table->entries[i];
747 if (pfn < PFN_UP(entry->addr))
748 register_nosave_region(pfn, PFN_UP(entry->addr));
750 pfn = PFN_DOWN(entry->addr + entry->size);
752 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
753 register_nosave_region(PFN_UP(entry->addr), pfn);
755 if (pfn >= limit_pfn)
762 * Register ACPI NVS memory regions, so that we can save/restore them during
763 * hibernation and the subsequent resume:
765 static int __init e820__register_nvs_regions(void)
769 for (i = 0; i < e820_table->nr_entries; i++) {
770 struct e820_entry *entry = &e820_table->entries[i];
772 if (entry->type == E820_TYPE_NVS)
773 acpi_nvs_register(entry->addr, entry->size);
778 core_initcall(e820__register_nvs_regions);
782 * Allocate the requested number of bytes with the requsted alignment
783 * and return (the physical address) to the caller. Also register this
784 * range in the 'kexec' E820 table as a reserved range.
786 * This allows kexec to fake a new mptable, as if it came from the real
789 u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
793 addr = memblock_phys_alloc(size, align);
795 e820__range_update_kexec(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
796 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
797 e820__update_table_kexec();
804 # ifdef CONFIG_X86_PAE
805 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
807 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
809 #else /* CONFIG_X86_32 */
810 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
814 * Find the highest page frame number we have available
816 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
819 unsigned long last_pfn = 0;
820 unsigned long max_arch_pfn = MAX_ARCH_PFN;
822 for (i = 0; i < e820_table->nr_entries; i++) {
823 struct e820_entry *entry = &e820_table->entries[i];
824 unsigned long start_pfn;
825 unsigned long end_pfn;
827 if (entry->type != type)
830 start_pfn = entry->addr >> PAGE_SHIFT;
831 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
833 if (start_pfn >= limit_pfn)
835 if (end_pfn > limit_pfn) {
836 last_pfn = limit_pfn;
839 if (end_pfn > last_pfn)
843 if (last_pfn > max_arch_pfn)
844 last_pfn = max_arch_pfn;
846 pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
847 last_pfn, max_arch_pfn);
851 unsigned long __init e820__end_of_ram_pfn(void)
853 return e820_end_pfn(MAX_ARCH_PFN, E820_TYPE_RAM);
856 unsigned long __init e820__end_of_low_ram_pfn(void)
858 return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
861 static void __init early_panic(char *msg)
867 static int userdef __initdata;
869 /* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
870 static int __init parse_memopt(char *p)
877 if (!strcmp(p, "nopentium")) {
879 setup_clear_cpu_cap(X86_FEATURE_PSE);
882 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
888 mem_size = memparse(p, &p);
890 /* Don't remove all memory when getting "mem={invalid}" parameter: */
894 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
896 #ifdef CONFIG_MEMORY_HOTPLUG
897 max_mem_size = mem_size;
902 early_param("mem", parse_memopt);
904 static int __init parse_memmap_one(char *p)
907 u64 start_at, mem_size;
912 if (!strncmp(p, "exactmap", 8)) {
913 e820_table->nr_entries = 0;
919 mem_size = memparse(p, &p);
925 start_at = memparse(p+1, &p);
926 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
927 } else if (*p == '#') {
928 start_at = memparse(p+1, &p);
929 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
930 } else if (*p == '$') {
931 start_at = memparse(p+1, &p);
932 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
933 } else if (*p == '!') {
934 start_at = memparse(p+1, &p);
935 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
936 } else if (*p == '%') {
937 enum e820_type from = 0, to = 0;
939 start_at = memparse(p + 1, &p);
941 from = simple_strtoull(p + 1, &p, 0);
943 to = simple_strtoull(p + 1, &p, 0);
947 e820__range_update(start_at, mem_size, from, to);
949 e820__range_add(start_at, mem_size, to);
951 e820__range_remove(start_at, mem_size, from, 1);
953 e820__range_remove(start_at, mem_size, 0, 0);
955 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
958 return *p == '\0' ? 0 : -EINVAL;
961 static int __init parse_memmap_opt(char *str)
964 char *k = strchr(str, ',');
969 parse_memmap_one(str);
975 early_param("memmap", parse_memmap_opt);
978 * Reserve all entries from the bootloader's extensible data nodes list,
979 * because if present we are going to use it later on to fetch e820
982 void __init e820__reserve_setup_data(void)
984 struct setup_data *data;
987 pa_data = boot_params.hdr.setup_data;
992 data = early_memremap(pa_data, sizeof(*data));
993 e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
994 e820__range_update_kexec(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
996 if (data->type == SETUP_INDIRECT &&
997 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT) {
998 e820__range_update(((struct setup_indirect *)data->data)->addr,
999 ((struct setup_indirect *)data->data)->len,
1000 E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1001 e820__range_update_kexec(((struct setup_indirect *)data->data)->addr,
1002 ((struct setup_indirect *)data->data)->len,
1003 E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1006 pa_data = data->next;
1007 early_memunmap(data, sizeof(*data));
1010 e820__update_table(e820_table);
1011 e820__update_table(e820_table_kexec);
1013 pr_info("extended physical RAM map:\n");
1014 e820__print_table("reserve setup_data");
1018 * Called after parse_early_param(), after early parameters (such as mem=)
1019 * have been processed, in which case we already have an E820 table filled in
1020 * via the parameter callback function(s), but it's not sorted and printed yet:
1022 void __init e820__finish_early_params(void)
1025 if (e820__update_table(e820_table) < 0)
1026 early_panic("Invalid user supplied memory map");
1028 pr_info("user-defined physical RAM map:\n");
1029 e820__print_table("user");
1033 static const char *__init e820_type_to_string(struct e820_entry *entry)
1035 switch (entry->type) {
1036 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1037 case E820_TYPE_RAM: return "System RAM";
1038 case E820_TYPE_ACPI: return "ACPI Tables";
1039 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1040 case E820_TYPE_UNUSABLE: return "Unusable memory";
1041 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1042 case E820_TYPE_PMEM: return "Persistent Memory";
1043 case E820_TYPE_RESERVED: return "Reserved";
1044 case E820_TYPE_SOFT_RESERVED: return "Soft Reserved";
1045 default: return "Unknown E820 type";
1049 static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1051 switch (entry->type) {
1052 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1053 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1054 case E820_TYPE_ACPI: /* Fall-through: */
1055 case E820_TYPE_NVS: /* Fall-through: */
1056 case E820_TYPE_UNUSABLE: /* Fall-through: */
1057 case E820_TYPE_PRAM: /* Fall-through: */
1058 case E820_TYPE_PMEM: /* Fall-through: */
1059 case E820_TYPE_RESERVED: /* Fall-through: */
1060 case E820_TYPE_SOFT_RESERVED: /* Fall-through: */
1061 default: return IORESOURCE_MEM;
1065 static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1067 switch (entry->type) {
1068 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1069 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1070 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1071 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1072 case E820_TYPE_RESERVED: return IORES_DESC_RESERVED;
1073 case E820_TYPE_SOFT_RESERVED: return IORES_DESC_SOFT_RESERVED;
1074 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1075 case E820_TYPE_RAM: /* Fall-through: */
1076 case E820_TYPE_UNUSABLE: /* Fall-through: */
1077 default: return IORES_DESC_NONE;
1081 static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1083 /* this is the legacy bios/dos rom-shadow + mmio region */
1084 if (res->start < (1ULL<<20))
1088 * Treat persistent memory and other special memory ranges like
1089 * device memory, i.e. reserve it for exclusive use of a driver
1092 case E820_TYPE_RESERVED:
1093 case E820_TYPE_SOFT_RESERVED:
1094 case E820_TYPE_PRAM:
1095 case E820_TYPE_PMEM:
1097 case E820_TYPE_RESERVED_KERN:
1099 case E820_TYPE_ACPI:
1101 case E820_TYPE_UNUSABLE:
1108 * Mark E820 reserved areas as busy for the resource manager:
1111 static struct resource __initdata *e820_res;
1113 void __init e820__reserve_resources(void)
1116 struct resource *res;
1119 res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1122 panic("%s: Failed to allocate %zu bytes\n", __func__,
1123 sizeof(*res) * e820_table->nr_entries);
1126 for (i = 0; i < e820_table->nr_entries; i++) {
1127 struct e820_entry *entry = e820_table->entries + i;
1129 end = entry->addr + entry->size - 1;
1130 if (end != (resource_size_t)end) {
1134 res->start = entry->addr;
1136 res->name = e820_type_to_string(entry);
1137 res->flags = e820_type_to_iomem_type(entry);
1138 res->desc = e820_type_to_iores_desc(entry);
1141 * Don't register the region that could be conflicted with
1142 * PCI device BAR resources and insert them later in
1143 * pcibios_resource_survey():
1145 if (do_mark_busy(entry->type, res)) {
1146 res->flags |= IORESOURCE_BUSY;
1147 insert_resource(&iomem_resource, res);
1152 /* Expose the bootloader-provided memory layout to the sysfs. */
1153 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1154 struct e820_entry *entry = e820_table_firmware->entries + i;
1156 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1161 * How much should we pad the end of RAM, depending on where it is?
1163 static unsigned long __init ram_alignment(resource_size_t pos)
1165 unsigned long mb = pos >> 20;
1167 /* To 64kB in the first megabyte */
1171 /* To 1MB in the first 16MB */
1175 /* To 64MB for anything above that */
1176 return 64*1024*1024;
1179 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1181 void __init e820__reserve_resources_late(void)
1184 struct resource *res;
1187 for (i = 0; i < e820_table->nr_entries; i++) {
1188 if (!res->parent && res->end)
1189 insert_resource_expand_to_fit(&iomem_resource, res);
1194 * Try to bump up RAM regions to reasonable boundaries, to
1197 for (i = 0; i < e820_table->nr_entries; i++) {
1198 struct e820_entry *entry = &e820_table->entries[i];
1201 if (entry->type != E820_TYPE_RAM)
1204 start = entry->addr + entry->size;
1205 end = round_up(start, ram_alignment(start)) - 1;
1206 if (end > MAX_RESOURCE_SIZE)
1207 end = MAX_RESOURCE_SIZE;
1211 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1212 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1217 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1219 char *__init e820__memory_setup_default(void)
1221 char *who = "BIOS-e820";
1224 * Try to copy the BIOS-supplied E820-map.
1226 * Otherwise fake a memory map; one section from 0k->640k,
1227 * the next section from 1mb->appropriate_mem_k
1229 if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1232 /* Compare results from other methods and take the one that gives more RAM: */
1233 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1234 mem_size = boot_params.screen_info.ext_mem_k;
1237 mem_size = boot_params.alt_mem_k;
1241 e820_table->nr_entries = 0;
1242 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1243 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1246 /* We just appended a lot of ranges, sanitize the table: */
1247 e820__update_table(e820_table);
1253 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1254 * E820 map - with an optional platform quirk available for virtual platforms
1255 * to override this method of boot environment processing:
1257 void __init e820__memory_setup(void)
1261 /* This is a firmware interface ABI - make sure we don't break it: */
1262 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1264 who = x86_init.resources.memory_setup();
1266 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1267 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1269 pr_info("BIOS-provided physical RAM map:\n");
1270 e820__print_table(who);
1273 void __init e820__memblock_setup(void)
1279 * The bootstrap memblock region count maximum is 128 entries
1280 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1281 * than that - so allow memblock resizing.
1283 * This is safe, because this call happens pretty late during x86 setup,
1284 * so we know about reserved memory regions already. (This is important
1285 * so that memblock resizing does no stomp over reserved areas.)
1287 memblock_allow_resize();
1289 for (i = 0; i < e820_table->nr_entries; i++) {
1290 struct e820_entry *entry = &e820_table->entries[i];
1292 end = entry->addr + entry->size;
1293 if (end != (resource_size_t)end)
1296 if (entry->type == E820_TYPE_SOFT_RESERVED)
1297 memblock_reserve(entry->addr, entry->size);
1299 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1302 memblock_add(entry->addr, entry->size);
1305 /* Throw away partial pages: */
1306 memblock_trim_memory(PAGE_SIZE);
1308 memblock_dump_all();