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/memblock.h>
13 #include <linux/suspend.h>
14 #include <linux/acpi.h>
15 #include <linux/firmware-map.h>
16 #include <linux/sort.h>
17 #include <linux/memory_hotplug.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.
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;
69 EXPORT_SYMBOL(pci_mem_start);
73 * This function checks if any part of the range <start,end> is mapped
76 static bool _e820__mapped_any(struct e820_table *table,
77 u64 start, u64 end, enum e820_type type)
81 for (i = 0; i < table->nr_entries; i++) {
82 struct e820_entry *entry = &table->entries[i];
84 if (type && entry->type != type)
86 if (entry->addr >= end || entry->addr + entry->size <= start)
93 bool e820__mapped_raw_any(u64 start, u64 end, enum e820_type type)
95 return _e820__mapped_any(e820_table_firmware, start, end, type);
97 EXPORT_SYMBOL_GPL(e820__mapped_raw_any);
99 bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
101 return _e820__mapped_any(e820_table, start, end, type);
103 EXPORT_SYMBOL_GPL(e820__mapped_any);
106 * This function checks if the entire <start,end> range is mapped with 'type'.
108 * Note: this function only works correctly once the E820 table is sorted and
109 * not-overlapping (at least for the range specified), which is the case normally.
111 static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
116 for (i = 0; i < e820_table->nr_entries; i++) {
117 struct e820_entry *entry = &e820_table->entries[i];
119 if (type && entry->type != type)
122 /* Is the region (part) in overlap with the current region? */
123 if (entry->addr >= end || entry->addr + entry->size <= start)
127 * If the region is at the beginning of <start,end> we move
128 * 'start' to the end of the region since it's ok until there
130 if (entry->addr <= start)
131 start = entry->addr + entry->size;
134 * If 'start' is now at or beyond 'end', we're done, full
135 * coverage of the desired range exists:
145 * This function checks if the entire range <start,end> is mapped with type.
147 bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
149 return __e820__mapped_all(start, end, type);
153 * This function returns the type associated with the range <start,end>.
155 int e820__get_entry_type(u64 start, u64 end)
157 struct e820_entry *entry = __e820__mapped_all(start, end, 0);
159 return entry ? entry->type : -EINVAL;
163 * Add a memory region to the kernel E820 map.
165 static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
167 int x = table->nr_entries;
169 if (x >= ARRAY_SIZE(table->entries)) {
170 pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
171 start, start + size - 1);
175 table->entries[x].addr = start;
176 table->entries[x].size = size;
177 table->entries[x].type = type;
181 void __init e820__range_add(u64 start, u64 size, enum e820_type type)
183 __e820__range_add(e820_table, start, size, type);
186 static void __init e820_print_type(enum e820_type type)
189 case E820_TYPE_RAM: /* Fall through: */
190 case E820_TYPE_RESERVED_KERN: pr_cont("usable"); break;
191 case E820_TYPE_RESERVED: pr_cont("reserved"); break;
192 case E820_TYPE_ACPI: pr_cont("ACPI data"); break;
193 case E820_TYPE_NVS: pr_cont("ACPI NVS"); break;
194 case E820_TYPE_UNUSABLE: pr_cont("unusable"); break;
195 case E820_TYPE_PMEM: /* Fall through: */
196 case E820_TYPE_PRAM: pr_cont("persistent (type %u)", type); break;
197 default: pr_cont("type %u", type); break;
201 void __init e820__print_table(char *who)
205 for (i = 0; i < e820_table->nr_entries; i++) {
206 pr_info("%s: [mem %#018Lx-%#018Lx] ",
208 e820_table->entries[i].addr,
209 e820_table->entries[i].addr + e820_table->entries[i].size - 1);
211 e820_print_type(e820_table->entries[i].type);
217 * Sanitize an E820 map.
219 * Some E820 layouts include overlapping entries. The following
220 * replaces the original E820 map with a new one, removing overlaps,
221 * and resolving conflicting memory types in favor of highest
224 * The input parameter 'entries' points to an array of 'struct
225 * e820_entry' which on entry has elements in the range [0, *nr_entries)
226 * valid, and which has space for up to max_nr_entries entries.
227 * On return, the resulting sanitized E820 map entries will be in
228 * overwritten in the same location, starting at 'entries'.
230 * The integer pointed to by nr_entries must be valid on entry (the
231 * current number of valid entries located at 'entries'). If the
232 * sanitizing succeeds the *nr_entries will be updated with the new
233 * number of valid entries (something no more than max_nr_entries).
235 * The return value from e820__update_table() is zero if it
236 * successfully 'sanitized' the map entries passed in, and is -1
237 * if it did nothing, which can happen if either of (1) it was
238 * only passed one map entry, or (2) any of the input map entries
239 * were invalid (start + size < start, meaning that the size was
240 * so big the described memory range wrapped around through zero.)
242 * Visually we're performing the following
243 * (1,2,3,4 = memory types)...
245 * Sample memory map (w/overlaps):
246 * ____22__________________
247 * ______________________4_
248 * ____1111________________
249 * _44_____________________
250 * 11111111________________
251 * ____________________33__
252 * ___________44___________
253 * __________33333_________
254 * ______________22________
255 * ___________________2222_
256 * _________111111111______
257 * _____________________11_
258 * _________________4______
260 * Sanitized equivalent (no overlap):
261 * 1_______________________
262 * _44_____________________
263 * ___1____________________
264 * ____22__________________
265 * ______11________________
266 * _________1______________
267 * __________3_____________
268 * ___________44___________
269 * _____________33_________
270 * _______________2________
271 * ________________1_______
272 * _________________4______
273 * ___________________2____
274 * ____________________33__
275 * ______________________4_
277 struct change_member {
278 /* Pointer to the original entry: */
279 struct e820_entry *entry;
280 /* Address for this change point: */
281 unsigned long long addr;
284 static struct change_member change_point_list[2*E820_MAX_ENTRIES] __initdata;
285 static struct change_member *change_point[2*E820_MAX_ENTRIES] __initdata;
286 static struct e820_entry *overlap_list[E820_MAX_ENTRIES] __initdata;
287 static struct e820_entry new_entries[E820_MAX_ENTRIES] __initdata;
289 static int __init cpcompare(const void *a, const void *b)
291 struct change_member * const *app = a, * const *bpp = b;
292 const struct change_member *ap = *app, *bp = *bpp;
295 * Inputs are pointers to two elements of change_point[]. If their
296 * addresses are not equal, their difference dominates. If the addresses
297 * are equal, then consider one that represents the end of its region
298 * to be greater than one that does not.
300 if (ap->addr != bp->addr)
301 return ap->addr > bp->addr ? 1 : -1;
303 return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
306 int __init e820__update_table(struct e820_table *table)
308 struct e820_entry *entries = table->entries;
309 u32 max_nr_entries = ARRAY_SIZE(table->entries);
310 enum e820_type current_type, last_type;
311 unsigned long long last_addr;
312 u32 new_nr_entries, overlap_entries;
313 u32 i, chg_idx, chg_nr;
315 /* If there's only one memory region, don't bother: */
316 if (table->nr_entries < 2)
319 BUG_ON(table->nr_entries > max_nr_entries);
321 /* Bail out if we find any unreasonable addresses in the map: */
322 for (i = 0; i < table->nr_entries; i++) {
323 if (entries[i].addr + entries[i].size < entries[i].addr)
327 /* Create pointers for initial change-point information (for sorting): */
328 for (i = 0; i < 2 * table->nr_entries; i++)
329 change_point[i] = &change_point_list[i];
332 * Record all known change-points (starting and ending addresses),
333 * omitting empty memory regions:
336 for (i = 0; i < table->nr_entries; i++) {
337 if (entries[i].size != 0) {
338 change_point[chg_idx]->addr = entries[i].addr;
339 change_point[chg_idx++]->entry = &entries[i];
340 change_point[chg_idx]->addr = entries[i].addr + entries[i].size;
341 change_point[chg_idx++]->entry = &entries[i];
346 /* Sort change-point list by memory addresses (low -> high): */
347 sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
349 /* Create a new memory map, removing overlaps: */
350 overlap_entries = 0; /* Number of entries in the overlap table */
351 new_nr_entries = 0; /* Index for creating new map entries */
352 last_type = 0; /* Start with undefined memory type */
353 last_addr = 0; /* Start with 0 as last starting address */
355 /* Loop through change-points, determining effect on the new map: */
356 for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
357 /* Keep track of all overlapping entries */
358 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
359 /* Add map entry to overlap list (> 1 entry implies an overlap) */
360 overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
362 /* Remove entry from list (order independent, so swap with last): */
363 for (i = 0; i < overlap_entries; i++) {
364 if (overlap_list[i] == change_point[chg_idx]->entry)
365 overlap_list[i] = overlap_list[overlap_entries-1];
370 * If there are overlapping entries, decide which
371 * "type" to use (larger value takes precedence --
372 * 1=usable, 2,3,4,4+=unusable)
375 for (i = 0; i < overlap_entries; i++) {
376 if (overlap_list[i]->type > current_type)
377 current_type = overlap_list[i]->type;
380 /* Continue building up new map based on this information: */
381 if (current_type != last_type || current_type == E820_TYPE_PRAM) {
382 if (last_type != 0) {
383 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
384 /* Move forward only if the new size was non-zero: */
385 if (new_entries[new_nr_entries].size != 0)
386 /* No more space left for new entries? */
387 if (++new_nr_entries >= max_nr_entries)
390 if (current_type != 0) {
391 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
392 new_entries[new_nr_entries].type = current_type;
393 last_addr = change_point[chg_idx]->addr;
395 last_type = current_type;
399 /* Copy the new entries into the original location: */
400 memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
401 table->nr_entries = new_nr_entries;
406 static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
408 struct boot_e820_entry *entry = entries;
411 u64 start = entry->addr;
412 u64 size = entry->size;
413 u64 end = start + size - 1;
414 u32 type = entry->type;
416 /* Ignore the entry on 64-bit overflow: */
417 if (start > end && likely(size))
420 e820__range_add(start, size, type);
429 * Copy the BIOS E820 map into a safe place.
431 * Sanity-check it while we're at it..
433 * If we're lucky and live on a modern system, the setup code
434 * will have given us a memory map that we can use to properly
435 * set up memory. If we aren't, we'll fake a memory map.
437 static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
439 /* Only one memory region (or negative)? Ignore it */
443 return __append_e820_table(entries, nr_entries);
447 __e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
451 u64 real_updated_size = 0;
453 BUG_ON(old_type == new_type);
455 if (size > (ULLONG_MAX - start))
456 size = ULLONG_MAX - start;
459 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
460 e820_print_type(old_type);
462 e820_print_type(new_type);
465 for (i = 0; i < table->nr_entries; i++) {
466 struct e820_entry *entry = &table->entries[i];
467 u64 final_start, final_end;
470 if (entry->type != old_type)
473 entry_end = entry->addr + entry->size;
475 /* Completely covered by new range? */
476 if (entry->addr >= start && entry_end <= end) {
477 entry->type = new_type;
478 real_updated_size += entry->size;
482 /* New range is completely covered? */
483 if (entry->addr < start && entry_end > end) {
484 __e820__range_add(table, start, size, new_type);
485 __e820__range_add(table, end, entry_end - end, entry->type);
486 entry->size = start - entry->addr;
487 real_updated_size += size;
491 /* Partially covered: */
492 final_start = max(start, entry->addr);
493 final_end = min(end, entry_end);
494 if (final_start >= final_end)
497 __e820__range_add(table, final_start, final_end - final_start, new_type);
499 real_updated_size += final_end - final_start;
502 * Left range could be head or tail, so need to update
505 entry->size -= final_end - final_start;
506 if (entry->addr < final_start)
509 entry->addr = final_end;
511 return real_updated_size;
514 u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
516 return __e820__range_update(e820_table, start, size, old_type, new_type);
519 static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
521 return __e820__range_update(e820_table_kexec, start, size, old_type, new_type);
524 /* Remove a range of memory from the E820 table: */
525 u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
529 u64 real_removed_size = 0;
531 if (size > (ULLONG_MAX - start))
532 size = ULLONG_MAX - start;
535 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
537 e820_print_type(old_type);
540 for (i = 0; i < e820_table->nr_entries; i++) {
541 struct e820_entry *entry = &e820_table->entries[i];
542 u64 final_start, final_end;
545 if (check_type && entry->type != old_type)
548 entry_end = entry->addr + entry->size;
550 /* Completely covered? */
551 if (entry->addr >= start && entry_end <= end) {
552 real_removed_size += entry->size;
553 memset(entry, 0, sizeof(*entry));
557 /* Is the new range completely covered? */
558 if (entry->addr < start && entry_end > end) {
559 e820__range_add(end, entry_end - end, entry->type);
560 entry->size = start - entry->addr;
561 real_removed_size += size;
565 /* Partially covered: */
566 final_start = max(start, entry->addr);
567 final_end = min(end, entry_end);
568 if (final_start >= final_end)
571 real_removed_size += final_end - final_start;
574 * Left range could be head or tail, so need to update
577 entry->size -= final_end - final_start;
578 if (entry->addr < final_start)
581 entry->addr = final_end;
583 return real_removed_size;
586 void __init e820__update_table_print(void)
588 if (e820__update_table(e820_table))
591 pr_info("modified physical RAM map:\n");
592 e820__print_table("modified");
595 static void __init e820__update_table_kexec(void)
597 e820__update_table(e820_table_kexec);
600 #define MAX_GAP_END 0x100000000ull
603 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
605 static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
607 unsigned long long last = MAX_GAP_END;
608 int i = e820_table->nr_entries;
612 unsigned long long start = e820_table->entries[i].addr;
613 unsigned long long end = start + e820_table->entries[i].size;
616 * Since "last" is at most 4GB, we know we'll
617 * fit in 32 bits if this condition is true:
620 unsigned long gap = last - end;
622 if (gap >= *gapsize) {
635 * Search for the biggest gap in the low 32 bits of the E820
636 * memory space. We pass this space to the PCI subsystem, so
637 * that it can assign MMIO resources for hotplug or
638 * unconfigured devices in.
640 * Hopefully the BIOS let enough space left.
642 __init void e820__setup_pci_gap(void)
644 unsigned long gapstart, gapsize;
648 found = e820_search_gap(&gapstart, &gapsize);
652 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
653 pr_err("Cannot find an available gap in the 32-bit address range\n");
654 pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
656 gapstart = 0x10000000;
661 * e820__reserve_resources_late() protects stolen RAM already:
663 pci_mem_start = gapstart;
665 pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
666 gapstart, gapstart + gapsize - 1);
670 * Called late during init, in free_initmem().
672 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
674 * Copy them to a (usually much smaller) dynamically allocated area that is
675 * sized precisely after the number of e820 entries.
677 * This is done after we've performed all the fixes and tweaks to the tables.
678 * All functions which modify them are __init functions, which won't exist
679 * after free_initmem().
681 __init void e820__reallocate_tables(void)
683 struct e820_table *n;
686 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
687 n = kmemdup(e820_table, size, GFP_KERNEL);
691 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
692 n = kmemdup(e820_table_kexec, size, GFP_KERNEL);
694 e820_table_kexec = n;
696 size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
697 n = kmemdup(e820_table_firmware, size, GFP_KERNEL);
699 e820_table_firmware = n;
703 * Because of the small fixed size of struct boot_params, only the first
704 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
705 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
706 * struct setup_data, which is parsed here.
708 void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
711 struct boot_e820_entry *extmap;
712 struct setup_data *sdata;
714 sdata = early_memremap(phys_addr, data_len);
715 entries = sdata->len / sizeof(*extmap);
716 extmap = (struct boot_e820_entry *)(sdata->data);
718 __append_e820_table(extmap, entries);
719 e820__update_table(e820_table);
721 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
722 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
724 early_memunmap(sdata, data_len);
725 pr_info("extended physical RAM map:\n");
726 e820__print_table("extended");
730 * Find the ranges of physical addresses that do not correspond to
731 * E820 RAM areas and register the corresponding pages as 'nosave' for
732 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
734 * This function requires the E820 map to be sorted and without any
735 * overlapping entries.
737 void __init e820__register_nosave_regions(unsigned long limit_pfn)
740 unsigned long pfn = 0;
742 for (i = 0; i < e820_table->nr_entries; i++) {
743 struct e820_entry *entry = &e820_table->entries[i];
745 if (pfn < PFN_UP(entry->addr))
746 register_nosave_region(pfn, PFN_UP(entry->addr));
748 pfn = PFN_DOWN(entry->addr + entry->size);
750 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
751 register_nosave_region(PFN_UP(entry->addr), pfn);
753 if (pfn >= limit_pfn)
760 * Register ACPI NVS memory regions, so that we can save/restore them during
761 * hibernation and the subsequent resume:
763 static int __init e820__register_nvs_regions(void)
767 for (i = 0; i < e820_table->nr_entries; i++) {
768 struct e820_entry *entry = &e820_table->entries[i];
770 if (entry->type == E820_TYPE_NVS)
771 acpi_nvs_register(entry->addr, entry->size);
776 core_initcall(e820__register_nvs_regions);
780 * Allocate the requested number of bytes with the requsted alignment
781 * and return (the physical address) to the caller. Also register this
782 * range in the 'kexec' E820 table as a reserved range.
784 * This allows kexec to fake a new mptable, as if it came from the real
787 u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
791 addr = memblock_phys_alloc(size, align);
793 e820__range_update_kexec(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
794 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
795 e820__update_table_kexec();
802 # ifdef CONFIG_X86_PAE
803 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
805 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
807 #else /* CONFIG_X86_32 */
808 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
812 * Find the highest page frame number we have available
814 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
817 unsigned long last_pfn = 0;
818 unsigned long max_arch_pfn = MAX_ARCH_PFN;
820 for (i = 0; i < e820_table->nr_entries; i++) {
821 struct e820_entry *entry = &e820_table->entries[i];
822 unsigned long start_pfn;
823 unsigned long end_pfn;
825 if (entry->type != type)
828 start_pfn = entry->addr >> PAGE_SHIFT;
829 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
831 if (start_pfn >= limit_pfn)
833 if (end_pfn > limit_pfn) {
834 last_pfn = limit_pfn;
837 if (end_pfn > last_pfn)
841 if (last_pfn > max_arch_pfn)
842 last_pfn = max_arch_pfn;
844 pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
845 last_pfn, max_arch_pfn);
849 unsigned long __init e820__end_of_ram_pfn(void)
851 return e820_end_pfn(MAX_ARCH_PFN, E820_TYPE_RAM);
854 unsigned long __init e820__end_of_low_ram_pfn(void)
856 return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
859 static void __init early_panic(char *msg)
865 static int userdef __initdata;
867 /* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
868 static int __init parse_memopt(char *p)
875 if (!strcmp(p, "nopentium")) {
877 setup_clear_cpu_cap(X86_FEATURE_PSE);
880 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
886 mem_size = memparse(p, &p);
888 /* Don't remove all memory when getting "mem={invalid}" parameter: */
892 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
894 #ifdef CONFIG_MEMORY_HOTPLUG
895 max_mem_size = mem_size;
900 early_param("mem", parse_memopt);
902 static int __init parse_memmap_one(char *p)
905 u64 start_at, mem_size;
910 if (!strncmp(p, "exactmap", 8)) {
911 #ifdef CONFIG_CRASH_DUMP
913 * If we are doing a crash dump, we still need to know
914 * the real memory size before the original memory map is
917 saved_max_pfn = e820__end_of_ram_pfn();
919 e820_table->nr_entries = 0;
925 mem_size = memparse(p, &p);
931 start_at = memparse(p+1, &p);
932 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
933 } else if (*p == '#') {
934 start_at = memparse(p+1, &p);
935 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
936 } else if (*p == '$') {
937 start_at = memparse(p+1, &p);
938 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
939 } else if (*p == '!') {
940 start_at = memparse(p+1, &p);
941 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
942 } else if (*p == '%') {
943 enum e820_type from = 0, to = 0;
945 start_at = memparse(p + 1, &p);
947 from = simple_strtoull(p + 1, &p, 0);
949 to = simple_strtoull(p + 1, &p, 0);
953 e820__range_update(start_at, mem_size, from, to);
955 e820__range_add(start_at, mem_size, to);
957 e820__range_remove(start_at, mem_size, from, 1);
959 e820__range_remove(start_at, mem_size, 0, 0);
961 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
964 return *p == '\0' ? 0 : -EINVAL;
967 static int __init parse_memmap_opt(char *str)
970 char *k = strchr(str, ',');
975 parse_memmap_one(str);
981 early_param("memmap", parse_memmap_opt);
984 * Reserve all entries from the bootloader's extensible data nodes list,
985 * because if present we are going to use it later on to fetch e820
988 void __init e820__reserve_setup_data(void)
990 struct setup_data *data;
993 pa_data = boot_params.hdr.setup_data;
998 data = early_memremap(pa_data, sizeof(*data));
999 e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1000 e820__range_update_kexec(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1001 pa_data = data->next;
1002 early_memunmap(data, sizeof(*data));
1005 e820__update_table(e820_table);
1006 e820__update_table(e820_table_kexec);
1008 pr_info("extended physical RAM map:\n");
1009 e820__print_table("reserve setup_data");
1013 * Called after parse_early_param(), after early parameters (such as mem=)
1014 * have been processed, in which case we already have an E820 table filled in
1015 * via the parameter callback function(s), but it's not sorted and printed yet:
1017 void __init e820__finish_early_params(void)
1020 if (e820__update_table(e820_table) < 0)
1021 early_panic("Invalid user supplied memory map");
1023 pr_info("user-defined physical RAM map:\n");
1024 e820__print_table("user");
1028 static const char *__init e820_type_to_string(struct e820_entry *entry)
1030 switch (entry->type) {
1031 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1032 case E820_TYPE_RAM: return "System RAM";
1033 case E820_TYPE_ACPI: return "ACPI Tables";
1034 case E820_TYPE_NVS: return "ACPI Non-volatile Storage";
1035 case E820_TYPE_UNUSABLE: return "Unusable memory";
1036 case E820_TYPE_PRAM: return "Persistent Memory (legacy)";
1037 case E820_TYPE_PMEM: return "Persistent Memory";
1038 case E820_TYPE_RESERVED: return "Reserved";
1039 default: return "Unknown E820 type";
1043 static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1045 switch (entry->type) {
1046 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1047 case E820_TYPE_RAM: return IORESOURCE_SYSTEM_RAM;
1048 case E820_TYPE_ACPI: /* Fall-through: */
1049 case E820_TYPE_NVS: /* Fall-through: */
1050 case E820_TYPE_UNUSABLE: /* Fall-through: */
1051 case E820_TYPE_PRAM: /* Fall-through: */
1052 case E820_TYPE_PMEM: /* Fall-through: */
1053 case E820_TYPE_RESERVED: /* Fall-through: */
1054 default: return IORESOURCE_MEM;
1058 static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1060 switch (entry->type) {
1061 case E820_TYPE_ACPI: return IORES_DESC_ACPI_TABLES;
1062 case E820_TYPE_NVS: return IORES_DESC_ACPI_NV_STORAGE;
1063 case E820_TYPE_PMEM: return IORES_DESC_PERSISTENT_MEMORY;
1064 case E820_TYPE_PRAM: return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1065 case E820_TYPE_RESERVED_KERN: /* Fall-through: */
1066 case E820_TYPE_RAM: /* Fall-through: */
1067 case E820_TYPE_UNUSABLE: /* Fall-through: */
1068 case E820_TYPE_RESERVED: /* Fall-through: */
1069 default: return IORES_DESC_NONE;
1073 static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1075 /* this is the legacy bios/dos rom-shadow + mmio region */
1076 if (res->start < (1ULL<<20))
1080 * Treat persistent memory like device memory, i.e. reserve it
1081 * for exclusive use of a driver
1084 case E820_TYPE_RESERVED:
1085 case E820_TYPE_PRAM:
1086 case E820_TYPE_PMEM:
1088 case E820_TYPE_RESERVED_KERN:
1090 case E820_TYPE_ACPI:
1092 case E820_TYPE_UNUSABLE:
1099 * Mark E820 reserved areas as busy for the resource manager:
1102 static struct resource __initdata *e820_res;
1104 void __init e820__reserve_resources(void)
1107 struct resource *res;
1110 res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1113 panic("%s: Failed to allocate %zu bytes\n", __func__,
1114 sizeof(*res) * e820_table->nr_entries);
1117 for (i = 0; i < e820_table->nr_entries; i++) {
1118 struct e820_entry *entry = e820_table->entries + i;
1120 end = entry->addr + entry->size - 1;
1121 if (end != (resource_size_t)end) {
1125 res->start = entry->addr;
1127 res->name = e820_type_to_string(entry);
1128 res->flags = e820_type_to_iomem_type(entry);
1129 res->desc = e820_type_to_iores_desc(entry);
1132 * Don't register the region that could be conflicted with
1133 * PCI device BAR resources and insert them later in
1134 * pcibios_resource_survey():
1136 if (do_mark_busy(entry->type, res)) {
1137 res->flags |= IORESOURCE_BUSY;
1138 insert_resource(&iomem_resource, res);
1143 /* Expose the bootloader-provided memory layout to the sysfs. */
1144 for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1145 struct e820_entry *entry = e820_table_firmware->entries + i;
1147 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1152 * How much should we pad the end of RAM, depending on where it is?
1154 static unsigned long __init ram_alignment(resource_size_t pos)
1156 unsigned long mb = pos >> 20;
1158 /* To 64kB in the first megabyte */
1162 /* To 1MB in the first 16MB */
1166 /* To 64MB for anything above that */
1167 return 64*1024*1024;
1170 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1172 void __init e820__reserve_resources_late(void)
1175 struct resource *res;
1178 for (i = 0; i < e820_table->nr_entries; i++) {
1179 if (!res->parent && res->end)
1180 insert_resource_expand_to_fit(&iomem_resource, res);
1185 * Try to bump up RAM regions to reasonable boundaries, to
1188 for (i = 0; i < e820_table->nr_entries; i++) {
1189 struct e820_entry *entry = &e820_table->entries[i];
1192 if (entry->type != E820_TYPE_RAM)
1195 start = entry->addr + entry->size;
1196 end = round_up(start, ram_alignment(start)) - 1;
1197 if (end > MAX_RESOURCE_SIZE)
1198 end = MAX_RESOURCE_SIZE;
1202 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1203 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1208 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1210 char *__init e820__memory_setup_default(void)
1212 char *who = "BIOS-e820";
1215 * Try to copy the BIOS-supplied E820-map.
1217 * Otherwise fake a memory map; one section from 0k->640k,
1218 * the next section from 1mb->appropriate_mem_k
1220 if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1223 /* Compare results from other methods and take the one that gives more RAM: */
1224 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1225 mem_size = boot_params.screen_info.ext_mem_k;
1228 mem_size = boot_params.alt_mem_k;
1232 e820_table->nr_entries = 0;
1233 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1234 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1237 /* We just appended a lot of ranges, sanitize the table: */
1238 e820__update_table(e820_table);
1244 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1245 * E820 map - with an optional platform quirk available for virtual platforms
1246 * to override this method of boot environment processing:
1248 void __init e820__memory_setup(void)
1252 /* This is a firmware interface ABI - make sure we don't break it: */
1253 BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1255 who = x86_init.resources.memory_setup();
1257 memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1258 memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1260 pr_info("BIOS-provided physical RAM map:\n");
1261 e820__print_table(who);
1264 void __init e820__memblock_setup(void)
1270 * The bootstrap memblock region count maximum is 128 entries
1271 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1272 * than that - so allow memblock resizing.
1274 * This is safe, because this call happens pretty late during x86 setup,
1275 * so we know about reserved memory regions already. (This is important
1276 * so that memblock resizing does no stomp over reserved areas.)
1278 memblock_allow_resize();
1280 for (i = 0; i < e820_table->nr_entries; i++) {
1281 struct e820_entry *entry = &e820_table->entries[i];
1283 end = entry->addr + entry->size;
1284 if (end != (resource_size_t)end)
1287 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1290 memblock_add(entry->addr, entry->size);
1293 /* Throw away partial pages: */
1294 memblock_trim_memory(PAGE_SIZE);
1296 memblock_dump_all();