e820.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Low level x86 E820 memory map handling functions.
   4 *
   5 * The firmware and bootloader passes us the "E820 table", which is the primary
   6 * physical memory layout description available about x86 systems.
   7 *
   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.).
  11 */
  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>
  19#include <linux/kvm_types.h>
  20
  21#include <asm/e820/api.h>
  22#include <asm/setup.h>
  23
  24/*
  25 * We organize the E820 table into three main data structures:
  26 *
  27 * - 'e820_table_firmware': the original firmware version passed to us by the
  28 *   bootloader - not modified by the kernel. It is composed of two parts:
  29 *   the first 128 E820 memory entries in boot_params.e820_table and the remaining
  30 *   (if any) entries of the SETUP_E820_EXT nodes. We use this to:
  31 *
  32 *       - the hibernation code uses it to generate a kernel-independent CRC32
  33 *         checksum of the physical memory layout of a system.
  34 *
  35 * - 'e820_table_kexec': a slightly modified (by the kernel) firmware version
  36 *   passed to us by the bootloader - the major difference between
  37 *   e820_table_firmware[] and this one is that e820_table_kexec[]
  38 *   might be modified by the kexec itself to fake an mptable.
  39 *   We use this to:
  40 *
  41 *       - kexec, which is a bootloader in disguise, uses the original E820
  42 *         layout to pass to the kexec-ed kernel. This way the original kernel
  43 *         can have a restricted E820 map while the kexec()-ed kexec-kernel
  44 *         can have access to full memory - etc.
  45 *
  46 *         Export the memory layout via /sys/firmware/memmap. kexec-tools uses
  47 *         the entries to create an E820 table for the kexec kernel.
  48 *
  49 *         kexec_file_load in-kernel code uses the table for the kexec kernel.
  50 *
  51 * - 'e820_table': this is the main E820 table that is massaged by the
  52 *   low level x86 platform code, or modified by boot parameters, before
  53 *   passed on to higher level MM layers.
  54 *
  55 * Once the E820 map has been converted to the standard Linux memory layout
  56 * information its role stops - modifying it has no effect and does not get
  57 * re-propagated. So its main role is a temporary bootstrap storage of firmware
  58 * specific memory layout data during early bootup.
  59 */
  60static struct e820_table e820_table_init		__initdata;
  61static struct e820_table e820_table_kexec_init		__initdata;
  62static struct e820_table e820_table_firmware_init	__initdata;
  63
  64struct e820_table *e820_table __refdata			= &e820_table_init;
  65struct e820_table *e820_table_kexec __refdata		= &e820_table_kexec_init;
  66struct e820_table *e820_table_firmware __refdata	= &e820_table_firmware_init;
  67
  68/* For PCI or other memory-mapped resources */
  69unsigned long pci_mem_start = 0xaeedbabe;
  70#ifdef CONFIG_PCI
  71EXPORT_SYMBOL(pci_mem_start);
  72#endif
  73
  74/*
  75 * This function checks if any part of the range <start,end> is mapped
  76 * with type.
  77 */
  78static bool _e820__mapped_any(struct e820_table *table,
  79			      u64 start, u64 end, enum e820_type type)
  80{
  81	int i;
  82
  83	for (i = 0; i < table->nr_entries; i++) {
  84		struct e820_entry *entry = &table->entries[i];
  85
  86		if (type && entry->type != type)
  87			continue;
  88		if (entry->addr >= end || entry->addr + entry->size <= start)
  89			continue;
  90		return true;
  91	}
  92	return false;
  93}
  94
  95bool e820__mapped_raw_any(u64 start, u64 end, enum e820_type type)
  96{
  97	return _e820__mapped_any(e820_table_firmware, start, end, type);
  98}
  99EXPORT_SYMBOL_FOR_KVM(e820__mapped_raw_any);
 100
 101bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
 102{
 103	return _e820__mapped_any(e820_table, start, end, type);
 104}
 105EXPORT_SYMBOL_GPL(e820__mapped_any);
 106
 107/*
 108 * This function checks if the entire <start,end> range is mapped with 'type'.
 109 *
 110 * Note: this function only works correctly once the E820 table is sorted and
 111 * not-overlapping (at least for the range specified), which is the case normally.
 112 */
 113static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
 114					     enum e820_type type)
 115{
 116	int i;
 117
 118	for (i = 0; i < e820_table->nr_entries; i++) {
 119		struct e820_entry *entry = &e820_table->entries[i];
 120
 121		if (type && entry->type != type)
 122			continue;
 123
 124		/* Is the region (part) in overlap with the current region? */
 125		if (entry->addr >= end || entry->addr + entry->size <= start)
 126			continue;
 127
 128		/*
 129		 * If the region is at the beginning of <start,end> we move
 130		 * 'start' to the end of the region since it's ok until there
 131		 */
 132		if (entry->addr <= start)
 133			start = entry->addr + entry->size;
 134
 135		/*
 136		 * If 'start' is now at or beyond 'end', we're done, full
 137		 * coverage of the desired range exists:
 138		 */
 139		if (start >= end)
 140			return entry;
 141	}
 142
 143	return NULL;
 144}
 145
 146/*
 147 * This function checks if the entire range <start,end> is mapped with type.
 148 */
 149bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
 150{
 151	return __e820__mapped_all(start, end, type);
 152}
 153
 154/*
 155 * This function returns the type associated with the range <start,end>.
 156 */
 157int e820__get_entry_type(u64 start, u64 end)
 158{
 159	struct e820_entry *entry = __e820__mapped_all(start, end, 0);
 160
 161	return entry ? entry->type : -EINVAL;
 162}
 163
 164/*
 165 * Add a memory region to the kernel E820 map.
 166 */
 167static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
 168{
 169	int x = table->nr_entries;
 170
 171	if (x >= ARRAY_SIZE(table->entries)) {
 172		pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
 173		       start, start + size - 1);
 174		return;
 175	}
 176
 177	table->entries[x].addr = start;
 178	table->entries[x].size = size;
 179	table->entries[x].type = type;
 180	table->nr_entries++;
 181}
 182
 183void __init e820__range_add(u64 start, u64 size, enum e820_type type)
 184{
 185	__e820__range_add(e820_table, start, size, type);
 186}
 187
 188static void __init e820_print_type(enum e820_type type)
 189{
 190	switch (type) {
 191	case E820_TYPE_RAM:		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;
 200	}
 201}
 202
 203void __init e820__print_table(char *who)
 204{
 205	int i;
 206
 207	for (i = 0; i < e820_table->nr_entries; i++) {
 208		pr_info("%s: [mem %#018Lx-%#018Lx] ",
 209			who,
 210			e820_table->entries[i].addr,
 211			e820_table->entries[i].addr + e820_table->entries[i].size - 1);
 212
 213		e820_print_type(e820_table->entries[i].type);
 214		pr_cont("\n");
 215	}
 216}
 217
 218/*
 219 * Sanitize an E820 map.
 220 *
 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
 224 * numbered type.
 225 *
 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'.
 231 *
 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).
 236 *
 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.)
 243 *
 244 *	Visually we're performing the following
 245 *	(1,2,3,4 = memory types)...
 246 *
 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______
 261 *
 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_
 278 */
 279struct change_member {
 280	/* Pointer to the original entry: */
 281	struct e820_entry	*entry;
 282	/* Address for this change point: */
 283	unsigned long long	addr;
 284};
 285
 286static struct change_member	change_point_list[2*E820_MAX_ENTRIES]	__initdata;
 287static struct change_member	*change_point[2*E820_MAX_ENTRIES]	__initdata;
 288static struct e820_entry	*overlap_list[E820_MAX_ENTRIES]		__initdata;
 289static struct e820_entry	new_entries[E820_MAX_ENTRIES]		__initdata;
 290
 291static int __init cpcompare(const void *a, const void *b)
 292{
 293	struct change_member * const *app = a, * const *bpp = b;
 294	const struct change_member *ap = *app, *bp = *bpp;
 295
 296	/*
 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.
 301	 */
 302	if (ap->addr != bp->addr)
 303		return ap->addr > bp->addr ? 1 : -1;
 304
 305	return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
 306}
 307
 308static bool e820_nomerge(enum e820_type type)
 309{
 310	/*
 311	 * These types may indicate distinct platform ranges aligned to
 312	 * numa node, protection domain, performance domain, or other
 313	 * boundaries. Do not merge them.
 314	 */
 315	if (type == E820_TYPE_PRAM)
 316		return true;
 317	if (type == E820_TYPE_SOFT_RESERVED)
 318		return true;
 319	return false;
 320}
 321
 322int __init e820__update_table(struct e820_table *table)
 323{
 324	struct e820_entry *entries = table->entries;
 325	u32 max_nr_entries = ARRAY_SIZE(table->entries);
 326	enum e820_type current_type, last_type;
 327	unsigned long long last_addr;
 328	u32 new_nr_entries, overlap_entries;
 329	u32 i, chg_idx, chg_nr;
 330
 331	/* If there's only one memory region, don't bother: */
 332	if (table->nr_entries < 2)
 333		return -1;
 334
 335	BUG_ON(table->nr_entries > max_nr_entries);
 336
 337	/* Bail out if we find any unreasonable addresses in the map: */
 338	for (i = 0; i < table->nr_entries; i++) {
 339		if (entries[i].addr + entries[i].size < entries[i].addr)
 340			return -1;
 341	}
 342
 343	/* Create pointers for initial change-point information (for sorting): */
 344	for (i = 0; i < 2 * table->nr_entries; i++)
 345		change_point[i] = &change_point_list[i];
 346
 347	/*
 348	 * Record all known change-points (starting and ending addresses),
 349	 * omitting empty memory regions:
 350	 */
 351	chg_idx = 0;
 352	for (i = 0; i < table->nr_entries; i++)	{
 353		if (entries[i].size != 0) {
 354			change_point[chg_idx]->addr	= entries[i].addr;
 355			change_point[chg_idx++]->entry	= &entries[i];
 356			change_point[chg_idx]->addr	= entries[i].addr + entries[i].size;
 357			change_point[chg_idx++]->entry	= &entries[i];
 358		}
 359	}
 360	chg_nr = chg_idx;
 361
 362	/* Sort change-point list by memory addresses (low -> high): */
 363	sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
 364
 365	/* Create a new memory map, removing overlaps: */
 366	overlap_entries = 0;	 /* Number of entries in the overlap table */
 367	new_nr_entries = 0;	 /* Index for creating new map entries */
 368	last_type = 0;		 /* Start with undefined memory type */
 369	last_addr = 0;		 /* Start with 0 as last starting address */
 370
 371	/* Loop through change-points, determining effect on the new map: */
 372	for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
 373		/* Keep track of all overlapping entries */
 374		if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
 375			/* Add map entry to overlap list (> 1 entry implies an overlap) */
 376			overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
 377		} else {
 378			/* Remove entry from list (order independent, so swap with last): */
 379			for (i = 0; i < overlap_entries; i++) {
 380				if (overlap_list[i] == change_point[chg_idx]->entry)
 381					overlap_list[i] = overlap_list[overlap_entries-1];
 382			}
 383			overlap_entries--;
 384		}
 385		/*
 386		 * If there are overlapping entries, decide which
 387		 * "type" to use (larger value takes precedence --
 388		 * 1=usable, 2,3,4,4+=unusable)
 389		 */
 390		current_type = 0;
 391		for (i = 0; i < overlap_entries; i++) {
 392			if (overlap_list[i]->type > current_type)
 393				current_type = overlap_list[i]->type;
 394		}
 395
 396		/* Continue building up new map based on this information: */
 397		if (current_type != last_type || e820_nomerge(current_type)) {
 398			if (last_type) {
 399				new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
 400				/* Move forward only if the new size was non-zero: */
 401				if (new_entries[new_nr_entries].size != 0)
 402					/* No more space left for new entries? */
 403					if (++new_nr_entries >= max_nr_entries)
 404						break;
 405			}
 406			if (current_type) {
 407				new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
 408				new_entries[new_nr_entries].type = current_type;
 409				last_addr = change_point[chg_idx]->addr;
 410			}
 411			last_type = current_type;
 412		}
 413	}
 414
 415	/* Copy the new entries into the original location: */
 416	memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
 417	table->nr_entries = new_nr_entries;
 418
 419	return 0;
 420}
 421
 422static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
 423{
 424	struct boot_e820_entry *entry = entries;
 425
 426	while (nr_entries) {
 427		u64 start = entry->addr;
 428		u64 size = entry->size;
 429		u64 end = start + size - 1;
 430		u32 type = entry->type;
 431
 432		/* Ignore the entry on 64-bit overflow: */
 433		if (start > end && likely(size))
 434			return -1;
 435
 436		e820__range_add(start, size, type);
 437
 438		entry++;
 439		nr_entries--;
 440	}
 441	return 0;
 442}
 443
 444/*
 445 * Copy the BIOS E820 map into a safe place.
 446 *
 447 * Sanity-check it while we're at it..
 448 *
 449 * If we're lucky and live on a modern system, the setup code
 450 * will have given us a memory map that we can use to properly
 451 * set up memory.  If we aren't, we'll fake a memory map.
 452 */
 453static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
 454{
 455	/* Only one memory region (or negative)? Ignore it */
 456	if (nr_entries < 2)
 457		return -1;
 458
 459	return __append_e820_table(entries, nr_entries);
 460}
 461
 462static u64 __init
 463__e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
 464{
 465	u64 end;
 466	unsigned int i;
 467	u64 real_updated_size = 0;
 468
 469	BUG_ON(old_type == new_type);
 470
 471	if (size > (ULLONG_MAX - start))
 472		size = ULLONG_MAX - start;
 473
 474	end = start + size;
 475	printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
 476	e820_print_type(old_type);
 477	pr_cont(" ==> ");
 478	e820_print_type(new_type);
 479	pr_cont("\n");
 480
 481	for (i = 0; i < table->nr_entries; i++) {
 482		struct e820_entry *entry = &table->entries[i];
 483		u64 final_start, final_end;
 484		u64 entry_end;
 485
 486		if (entry->type != old_type)
 487			continue;
 488
 489		entry_end = entry->addr + entry->size;
 490
 491		/* Completely covered by new range? */
 492		if (entry->addr >= start && entry_end <= end) {
 493			entry->type = new_type;
 494			real_updated_size += entry->size;
 495			continue;
 496		}
 497
 498		/* New range is completely covered? */
 499		if (entry->addr < start && entry_end > end) {
 500			__e820__range_add(table, start, size, new_type);
 501			__e820__range_add(table, end, entry_end - end, entry->type);
 502			entry->size = start - entry->addr;
 503			real_updated_size += size;
 504			continue;
 505		}
 506
 507		/* Partially covered: */
 508		final_start = max(start, entry->addr);
 509		final_end = min(end, entry_end);
 510		if (final_start >= final_end)
 511			continue;
 512
 513		__e820__range_add(table, final_start, final_end - final_start, new_type);
 514
 515		real_updated_size += final_end - final_start;
 516
 517		/*
 518		 * Left range could be head or tail, so need to update
 519		 * its size first:
 520		 */
 521		entry->size -= final_end - final_start;
 522		if (entry->addr < final_start)
 523			continue;
 524
 525		entry->addr = final_end;
 526	}
 527	return real_updated_size;
 528}
 529
 530u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
 531{
 532	return __e820__range_update(e820_table, start, size, old_type, new_type);
 533}
 534
 535u64 __init e820__range_update_table(struct e820_table *t, u64 start, u64 size,
 536				    enum e820_type old_type, enum e820_type new_type)
 537{
 538	return __e820__range_update(t, start, size, old_type, new_type);
 539}
 540
 541/* Remove a range of memory from the E820 table: */
 542u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
 543{
 544	int i;
 545	u64 end;
 546	u64 real_removed_size = 0;
 547
 548	if (size > (ULLONG_MAX - start))
 549		size = ULLONG_MAX - start;
 550
 551	end = start + size;
 552	printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
 553	if (check_type)
 554		e820_print_type(old_type);
 555	pr_cont("\n");
 556
 557	for (i = 0; i < e820_table->nr_entries; i++) {
 558		struct e820_entry *entry = &e820_table->entries[i];
 559		u64 final_start, final_end;
 560		u64 entry_end;
 561
 562		if (check_type && entry->type != old_type)
 563			continue;
 564
 565		entry_end = entry->addr + entry->size;
 566
 567		/* Completely covered? */
 568		if (entry->addr >= start && entry_end <= end) {
 569			real_removed_size += entry->size;
 570			memset(entry, 0, sizeof(*entry));
 571			continue;
 572		}
 573
 574		/* Is the new range completely covered? */
 575		if (entry->addr < start && entry_end > end) {
 576			e820__range_add(end, entry_end - end, entry->type);
 577			entry->size = start - entry->addr;
 578			real_removed_size += size;
 579			continue;
 580		}
 581
 582		/* Partially covered: */
 583		final_start = max(start, entry->addr);
 584		final_end = min(end, entry_end);
 585		if (final_start >= final_end)
 586			continue;
 587
 588		real_removed_size += final_end - final_start;
 589
 590		/*
 591		 * Left range could be head or tail, so need to update
 592		 * the size first:
 593		 */
 594		entry->size -= final_end - final_start;
 595		if (entry->addr < final_start)
 596			continue;
 597
 598		entry->addr = final_end;
 599	}
 600	return real_removed_size;
 601}
 602
 603void __init e820__update_table_print(void)
 604{
 605	if (e820__update_table(e820_table))
 606		return;
 607
 608	pr_info("modified physical RAM map:\n");
 609	e820__print_table("modified");
 610}
 611
 612static void __init e820__update_table_kexec(void)
 613{
 614	e820__update_table(e820_table_kexec);
 615}
 616
 617#define MAX_GAP_END 0x100000000ull
 618
 619/*
 620 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
 621 */
 622static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
 623{
 624	unsigned long long last = MAX_GAP_END;
 625	int i = e820_table->nr_entries;
 626	int found = 0;
 627
 628	while (--i >= 0) {
 629		unsigned long long start = e820_table->entries[i].addr;
 630		unsigned long long end = start + e820_table->entries[i].size;
 631
 632		/*
 633		 * Since "last" is at most 4GB, we know we'll
 634		 * fit in 32 bits if this condition is true:
 635		 */
 636		if (last > end) {
 637			unsigned long gap = last - end;
 638
 639			if (gap >= *gapsize) {
 640				*gapsize = gap;
 641				*gapstart = end;
 642				found = 1;
 643			}
 644		}
 645		if (start < last)
 646			last = start;
 647	}
 648	return found;
 649}
 650
 651/*
 652 * Search for the biggest gap in the low 32 bits of the E820
 653 * memory space. We pass this space to the PCI subsystem, so
 654 * that it can assign MMIO resources for hotplug or
 655 * unconfigured devices in.
 656 *
 657 * Hopefully the BIOS let enough space left.
 658 */
 659__init void e820__setup_pci_gap(void)
 660{
 661	unsigned long gapstart, gapsize;
 662	int found;
 663
 664	gapsize = 0x400000;
 665	found  = e820_search_gap(&gapstart, &gapsize);
 666
 667	if (!found) {
 668#ifdef CONFIG_X86_64
 669		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
 670		pr_err("Cannot find an available gap in the 32-bit address range\n");
 671		pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
 672#else
 673		gapstart = 0x10000000;
 674#endif
 675	}
 676
 677	/*
 678	 * e820__reserve_resources_late() protects stolen RAM already:
 679	 */
 680	pci_mem_start = gapstart;
 681
 682	pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
 683		gapstart, gapstart + gapsize - 1);
 684}
 685
 686/*
 687 * Called late during init, in free_initmem().
 688 *
 689 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
 690 *
 691 * Copy them to a (usually much smaller) dynamically allocated area that is
 692 * sized precisely after the number of e820 entries.
 693 *
 694 * This is done after we've performed all the fixes and tweaks to the tables.
 695 * All functions which modify them are __init functions, which won't exist
 696 * after free_initmem().
 697 */
 698__init void e820__reallocate_tables(void)
 699{
 700	struct e820_table *n;
 701	int size;
 702
 703	size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
 704	n = kmemdup(e820_table, size, GFP_KERNEL);
 705	BUG_ON(!n);
 706	e820_table = n;
 707
 708	size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
 709	n = kmemdup(e820_table_kexec, size, GFP_KERNEL);
 710	BUG_ON(!n);
 711	e820_table_kexec = n;
 712
 713	size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
 714	n = kmemdup(e820_table_firmware, size, GFP_KERNEL);
 715	BUG_ON(!n);
 716	e820_table_firmware = n;
 717}
 718
 719/*
 720 * Because of the small fixed size of struct boot_params, only the first
 721 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
 722 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
 723 * struct setup_data, which is parsed here.
 724 */
 725void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
 726{
 727	int entries;
 728	struct boot_e820_entry *extmap;
 729	struct setup_data *sdata;
 730
 731	sdata = early_memremap(phys_addr, data_len);
 732	entries = sdata->len / sizeof(*extmap);
 733	extmap = (struct boot_e820_entry *)(sdata->data);
 734
 735	__append_e820_table(extmap, entries);
 736	e820__update_table(e820_table);
 737
 738	memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
 739	memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
 740
 741	early_memunmap(sdata, data_len);
 742	pr_info("extended physical RAM map:\n");
 743	e820__print_table("extended");
 744}
 745
 746/*
 747 * Find the ranges of physical addresses that do not correspond to
 748 * E820 RAM areas and register the corresponding pages as 'nosave' for
 749 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
 750 *
 751 * This function requires the E820 map to be sorted and without any
 752 * overlapping entries.
 753 */
 754void __init e820__register_nosave_regions(unsigned long limit_pfn)
 755{
 756	int i;
 757	u64 last_addr = 0;
 758
 759	for (i = 0; i < e820_table->nr_entries; i++) {
 760		struct e820_entry *entry = &e820_table->entries[i];
 761
 762		if (entry->type != E820_TYPE_RAM)
 763			continue;
 764
 765		if (last_addr < entry->addr)
 766			register_nosave_region(PFN_DOWN(last_addr), PFN_UP(entry->addr));
 767
 768		last_addr = entry->addr + entry->size;
 769	}
 770
 771	register_nosave_region(PFN_DOWN(last_addr), limit_pfn);
 772}
 773
 774#ifdef CONFIG_ACPI
 775/*
 776 * Register ACPI NVS memory regions, so that we can save/restore them during
 777 * hibernation and the subsequent resume:
 778 */
 779static int __init e820__register_nvs_regions(void)
 780{
 781	int i;
 782
 783	for (i = 0; i < e820_table->nr_entries; i++) {
 784		struct e820_entry *entry = &e820_table->entries[i];
 785
 786		if (entry->type == E820_TYPE_NVS)
 787			acpi_nvs_register(entry->addr, entry->size);
 788	}
 789
 790	return 0;
 791}
 792core_initcall(e820__register_nvs_regions);
 793#endif
 794
 795/*
 796 * Allocate the requested number of bytes with the requested alignment
 797 * and return (the physical address) to the caller. Also register this
 798 * range in the 'kexec' E820 table as a reserved range.
 799 *
 800 * This allows kexec to fake a new mptable, as if it came from the real
 801 * system.
 802 */
 803u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
 804{
 805	u64 addr;
 806
 807	addr = memblock_phys_alloc(size, align);
 808	if (addr) {
 809		e820__range_update_table(e820_table_kexec, addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
 810		pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
 811		e820__update_table_kexec();
 812	}
 813
 814	return addr;
 815}
 816
 817#ifdef CONFIG_X86_32
 818# ifdef CONFIG_X86_PAE
 819#  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
 820# else
 821#  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
 822# endif
 823#else /* CONFIG_X86_32 */
 824# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
 825#endif
 826
 827/*
 828 * Find the highest page frame number we have available
 829 */
 830static unsigned long __init e820__end_ram_pfn(unsigned long limit_pfn)
 831{
 832	int i;
 833	unsigned long last_pfn = 0;
 834	unsigned long max_arch_pfn = MAX_ARCH_PFN;
 835
 836	for (i = 0; i < e820_table->nr_entries; i++) {
 837		struct e820_entry *entry = &e820_table->entries[i];
 838		unsigned long start_pfn;
 839		unsigned long end_pfn;
 840
 841		if (entry->type != E820_TYPE_RAM &&
 842		    entry->type != E820_TYPE_ACPI)
 843			continue;
 844
 845		start_pfn = entry->addr >> PAGE_SHIFT;
 846		end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
 847
 848		if (start_pfn >= limit_pfn)
 849			continue;
 850		if (end_pfn > limit_pfn) {
 851			last_pfn = limit_pfn;
 852			break;
 853		}
 854		if (end_pfn > last_pfn)
 855			last_pfn = end_pfn;
 856	}
 857
 858	if (last_pfn > max_arch_pfn)
 859		last_pfn = max_arch_pfn;
 860
 861	pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
 862		last_pfn, max_arch_pfn);
 863	return last_pfn;
 864}
 865
 866unsigned long __init e820__end_of_ram_pfn(void)
 867{
 868	return e820__end_ram_pfn(MAX_ARCH_PFN);
 869}
 870
 871unsigned long __init e820__end_of_low_ram_pfn(void)
 872{
 873	return e820__end_ram_pfn(1UL << (32 - PAGE_SHIFT));
 874}
 875
 876static void __init early_panic(char *msg)
 877{
 878	early_printk(msg);
 879	panic(msg);
 880}
 881
 882static int userdef __initdata;
 883
 884/* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
 885static int __init parse_memopt(char *p)
 886{
 887	u64 mem_size;
 888
 889	if (!p)
 890		return -EINVAL;
 891
 892	if (!strcmp(p, "nopentium")) {
 893#ifdef CONFIG_X86_32
 894		setup_clear_cpu_cap(X86_FEATURE_PSE);
 895		return 0;
 896#else
 897		pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
 898		return -EINVAL;
 899#endif
 900	}
 901
 902	userdef = 1;
 903	mem_size = memparse(p, &p);
 904
 905	/* Don't remove all memory when getting "mem={invalid}" parameter: */
 906	if (mem_size == 0)
 907		return -EINVAL;
 908
 909	e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
 910
 911#ifdef CONFIG_MEMORY_HOTPLUG
 912	max_mem_size = mem_size;
 913#endif
 914
 915	return 0;
 916}
 917early_param("mem", parse_memopt);
 918
 919static int __init parse_memmap_one(char *p)
 920{
 921	char *oldp;
 922	u64 start_at, mem_size;
 923
 924	if (!p)
 925		return -EINVAL;
 926
 927	if (!strncmp(p, "exactmap", 8)) {
 928		e820_table->nr_entries = 0;
 929		userdef = 1;
 930		return 0;
 931	}
 932
 933	oldp = p;
 934	mem_size = memparse(p, &p);
 935	if (p == oldp)
 936		return -EINVAL;
 937
 938	userdef = 1;
 939	if (*p == '@') {
 940		start_at = memparse(p+1, &p);
 941		e820__range_add(start_at, mem_size, E820_TYPE_RAM);
 942	} else if (*p == '#') {
 943		start_at = memparse(p+1, &p);
 944		e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
 945	} else if (*p == '$') {
 946		start_at = memparse(p+1, &p);
 947		e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
 948	} else if (*p == '!') {
 949		start_at = memparse(p+1, &p);
 950		e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
 951	} else if (*p == '%') {
 952		enum e820_type from = 0, to = 0;
 953
 954		start_at = memparse(p + 1, &p);
 955		if (*p == '-')
 956			from = simple_strtoull(p + 1, &p, 0);
 957		if (*p == '+')
 958			to = simple_strtoull(p + 1, &p, 0);
 959		if (*p != '\0')
 960			return -EINVAL;
 961		if (from && to)
 962			e820__range_update(start_at, mem_size, from, to);
 963		else if (to)
 964			e820__range_add(start_at, mem_size, to);
 965		else if (from)
 966			e820__range_remove(start_at, mem_size, from, 1);
 967		else
 968			e820__range_remove(start_at, mem_size, 0, 0);
 969	} else {
 970		e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
 971	}
 972
 973	return *p == '\0' ? 0 : -EINVAL;
 974}
 975
 976static int __init parse_memmap_opt(char *str)
 977{
 978	while (str) {
 979		char *k = strchr(str, ',');
 980
 981		if (k)
 982			*k++ = 0;
 983
 984		parse_memmap_one(str);
 985		str = k;
 986	}
 987
 988	return 0;
 989}
 990early_param("memmap", parse_memmap_opt);
 991
 992/*
 993 * Called after parse_early_param(), after early parameters (such as mem=)
 994 * have been processed, in which case we already have an E820 table filled in
 995 * via the parameter callback function(s), but it's not sorted and printed yet:
 996 */
 997void __init e820__finish_early_params(void)
 998{
 999	if (userdef) {
1000		if (e820__update_table(e820_table) < 0)
1001			early_panic("Invalid user supplied memory map");
1002
1003		pr_info("user-defined physical RAM map:\n");
1004		e820__print_table("user");
1005	}
1006}
1007
1008static const char *__init e820_type_to_string(struct e820_entry *entry)
1009{
1010	switch (entry->type) {
1011	case E820_TYPE_RAM:		return "System RAM";
1012	case E820_TYPE_ACPI:		return "ACPI Tables";
1013	case E820_TYPE_NVS:		return "ACPI Non-volatile Storage";
1014	case E820_TYPE_UNUSABLE:	return "Unusable memory";
1015	case E820_TYPE_PRAM:		return "Persistent Memory (legacy)";
1016	case E820_TYPE_PMEM:		return "Persistent Memory";
1017	case E820_TYPE_RESERVED:	return "Reserved";
1018	case E820_TYPE_SOFT_RESERVED:	return "Soft Reserved";
1019	default:			return "Unknown E820 type";
1020	}
1021}
1022
1023static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1024{
1025	switch (entry->type) {
1026	case E820_TYPE_RAM:		return IORESOURCE_SYSTEM_RAM;
1027	case E820_TYPE_ACPI:		/* Fall-through: */
1028	case E820_TYPE_NVS:		/* Fall-through: */
1029	case E820_TYPE_UNUSABLE:	/* Fall-through: */
1030	case E820_TYPE_PRAM:		/* Fall-through: */
1031	case E820_TYPE_PMEM:		/* Fall-through: */
1032	case E820_TYPE_RESERVED:	/* Fall-through: */
1033	case E820_TYPE_SOFT_RESERVED:	/* Fall-through: */
1034	default:			return IORESOURCE_MEM;
1035	}
1036}
1037
1038static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1039{
1040	switch (entry->type) {
1041	case E820_TYPE_ACPI:		return IORES_DESC_ACPI_TABLES;
1042	case E820_TYPE_NVS:		return IORES_DESC_ACPI_NV_STORAGE;
1043	case E820_TYPE_PMEM:		return IORES_DESC_PERSISTENT_MEMORY;
1044	case E820_TYPE_PRAM:		return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1045	case E820_TYPE_RESERVED:	return IORES_DESC_RESERVED;
1046	case E820_TYPE_SOFT_RESERVED:	return IORES_DESC_SOFT_RESERVED;
1047	case E820_TYPE_RAM:		/* Fall-through: */
1048	case E820_TYPE_UNUSABLE:	/* Fall-through: */
1049	default:			return IORES_DESC_NONE;
1050	}
1051}
1052
1053static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1054{
1055	/* this is the legacy bios/dos rom-shadow + mmio region */
1056	if (res->start < (1ULL<<20))
1057		return true;
1058
1059	/*
1060	 * Treat persistent memory and other special memory ranges like
1061	 * device memory, i.e. reserve it for exclusive use of a driver
1062	 */
1063	switch (type) {
1064	case E820_TYPE_RESERVED:
1065	case E820_TYPE_SOFT_RESERVED:
1066	case E820_TYPE_PRAM:
1067	case E820_TYPE_PMEM:
1068		return false;
1069	case E820_TYPE_RAM:
1070	case E820_TYPE_ACPI:
1071	case E820_TYPE_NVS:
1072	case E820_TYPE_UNUSABLE:
1073	default:
1074		return true;
1075	}
1076}
1077
1078/*
1079 * Mark E820 reserved areas as busy for the resource manager:
1080 */
1081
1082static struct resource __initdata *e820_res;
1083
1084void __init e820__reserve_resources(void)
1085{
1086	int i;
1087	struct resource *res;
1088	u64 end;
1089
1090	res = memblock_alloc_or_panic(sizeof(*res) * e820_table->nr_entries,
1091			     SMP_CACHE_BYTES);
1092	e820_res = res;
1093
1094	for (i = 0; i < e820_table->nr_entries; i++) {
1095		struct e820_entry *entry = e820_table->entries + i;
1096
1097		end = entry->addr + entry->size - 1;
1098		if (end != (resource_size_t)end) {
1099			res++;
1100			continue;
1101		}
1102		res->start = entry->addr;
1103		res->end   = end;
1104		res->name  = e820_type_to_string(entry);
1105		res->flags = e820_type_to_iomem_type(entry);
1106		res->desc  = e820_type_to_iores_desc(entry);
1107
1108		/*
1109		 * Don't register the region that could be conflicted with
1110		 * PCI device BAR resources and insert them later in
1111		 * pcibios_resource_survey():
1112		 */
1113		if (do_mark_busy(entry->type, res)) {
1114			res->flags |= IORESOURCE_BUSY;
1115			insert_resource(&iomem_resource, res);
1116		}
1117		res++;
1118	}
1119
1120	/* Expose the kexec e820 table to the sysfs. */
1121	for (i = 0; i < e820_table_kexec->nr_entries; i++) {
1122		struct e820_entry *entry = e820_table_kexec->entries + i;
1123
1124		firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1125	}
1126}
1127
1128/*
1129 * How much should we pad the end of RAM, depending on where it is?
1130 */
1131static unsigned long __init ram_alignment(resource_size_t pos)
1132{
1133	unsigned long mb = pos >> 20;
1134
1135	/* To 64kB in the first megabyte */
1136	if (!mb)
1137		return 64*1024;
1138
1139	/* To 1MB in the first 16MB */
1140	if (mb < 16)
1141		return 1024*1024;
1142
1143	/* To 64MB for anything above that */
1144	return 64*1024*1024;
1145}
1146
1147#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1148
1149void __init e820__reserve_resources_late(void)
1150{
1151	int i;
1152	struct resource *res;
1153
1154	res = e820_res;
1155	for (i = 0; i < e820_table->nr_entries; i++) {
1156		if (!res->parent && res->end)
1157			insert_resource_expand_to_fit(&iomem_resource, res);
1158		res++;
1159	}
1160
1161	/*
1162	 * Try to bump up RAM regions to reasonable boundaries, to
1163	 * avoid stolen RAM:
1164	 */
1165	for (i = 0; i < e820_table->nr_entries; i++) {
1166		struct e820_entry *entry = &e820_table->entries[i];
1167		u64 start, end;
1168
1169		if (entry->type != E820_TYPE_RAM)
1170			continue;
1171
1172		start = entry->addr + entry->size;
1173		end = round_up(start, ram_alignment(start)) - 1;
1174		if (end > MAX_RESOURCE_SIZE)
1175			end = MAX_RESOURCE_SIZE;
1176		if (start >= end)
1177			continue;
1178
1179		printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1180		reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1181	}
1182}
1183
1184/*
1185 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1186 */
1187char *__init e820__memory_setup_default(void)
1188{
1189	char *who = "BIOS-e820";
1190
1191	/*
1192	 * Try to copy the BIOS-supplied E820-map.
1193	 *
1194	 * Otherwise fake a memory map; one section from 0k->640k,
1195	 * the next section from 1mb->appropriate_mem_k
1196	 */
1197	if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1198		u64 mem_size;
1199
1200		/* Compare results from other methods and take the one that gives more RAM: */
1201		if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1202			mem_size = boot_params.screen_info.ext_mem_k;
1203			who = "BIOS-88";
1204		} else {
1205			mem_size = boot_params.alt_mem_k;
1206			who = "BIOS-e801";
1207		}
1208
1209		e820_table->nr_entries = 0;
1210		e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1211		e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1212	}
1213
1214	/* We just appended a lot of ranges, sanitize the table: */
1215	e820__update_table(e820_table);
1216
1217	return who;
1218}
1219
1220/*
1221 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1222 * E820 map - with an optional platform quirk available for virtual platforms
1223 * to override this method of boot environment processing:
1224 */
1225void __init e820__memory_setup(void)
1226{
1227	char *who;
1228
1229	/* This is a firmware interface ABI - make sure we don't break it: */
1230	BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1231
1232	who = x86_init.resources.memory_setup();
1233
1234	memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1235	memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1236
1237	pr_info("BIOS-provided physical RAM map:\n");
1238	e820__print_table(who);
1239}
1240
1241void __init e820__memblock_setup(void)
1242{
1243	int i;
1244	u64 end;
1245
1246#ifdef CONFIG_MEMORY_HOTPLUG
1247	/*
1248	 * Memory used by the kernel cannot be hot-removed because Linux
1249	 * cannot migrate the kernel pages. When memory hotplug is
1250	 * enabled, we should prevent memblock from allocating memory
1251	 * for the kernel.
1252	 *
1253	 * ACPI SRAT records all hotpluggable memory ranges. But before
1254	 * SRAT is parsed, we don't know about it.
1255	 *
1256	 * The kernel image is loaded into memory at very early time. We
1257	 * cannot prevent this anyway. So on NUMA system, we set any
1258	 * node the kernel resides in as un-hotpluggable.
1259	 *
1260	 * Since on modern servers, one node could have double-digit
1261	 * gigabytes memory, we can assume the memory around the kernel
1262	 * image is also un-hotpluggable. So before SRAT is parsed, just
1263	 * allocate memory near the kernel image to try the best to keep
1264	 * the kernel away from hotpluggable memory.
1265	 */
1266	if (movable_node_is_enabled())
1267		memblock_set_bottom_up(true);
1268#endif
1269
1270	/*
1271	 * At this point only the first megabyte is mapped for sure, the
1272	 * rest of the memory cannot be used for memblock resizing
1273	 */
1274	memblock_set_current_limit(ISA_END_ADDRESS);
1275
1276	/*
1277	 * The bootstrap memblock region count maximum is 128 entries
1278	 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1279	 * than that - so allow memblock resizing.
1280	 *
1281	 * This is safe, because this call happens pretty late during x86 setup,
1282	 * so we know about reserved memory regions already. (This is important
1283	 * so that memblock resizing does no stomp over reserved areas.)
1284	 */
1285	memblock_allow_resize();
1286
1287	for (i = 0; i < e820_table->nr_entries; i++) {
1288		struct e820_entry *entry = &e820_table->entries[i];
1289
1290		end = entry->addr + entry->size;
1291		if (end != (resource_size_t)end)
1292			continue;
1293
1294		if (entry->type == E820_TYPE_SOFT_RESERVED)
1295			memblock_reserve(entry->addr, entry->size);
1296
1297		if (entry->type != E820_TYPE_RAM)
1298			continue;
1299
1300		memblock_add(entry->addr, entry->size);
1301	}
1302
1303	/*
1304	 * At this point memblock is only allowed to allocate from memory
1305	 * below 1M (aka ISA_END_ADDRESS) up until direct map is completely set
1306	 * up in init_mem_mapping().
1307	 *
1308	 * KHO kernels are special and use only scratch memory for memblock
1309	 * allocations, but memory below 1M is ignored by kernel after early
1310	 * boot and cannot be naturally marked as scratch.
1311	 *
1312	 * To allow allocation of the real-mode trampoline and a few (if any)
1313	 * other very early allocations from below 1M forcibly mark the memory
1314	 * below 1M as scratch.
1315	 *
1316	 * After real mode trampoline is allocated, we clear that scratch
1317	 * marking.
1318	 */
1319	memblock_mark_kho_scratch(0, SZ_1M);
1320
1321	/*
1322	 * 32-bit systems are limited to 4BG of memory even with HIGHMEM and
1323	 * to even less without it.
1324	 * Discard memory after max_pfn - the actual limit detected at runtime.
1325	 */
1326	if (IS_ENABLED(CONFIG_X86_32))
1327		memblock_remove(PFN_PHYS(max_pfn), -1);
1328
1329	/* Throw away partial pages: */
1330	memblock_trim_memory(PAGE_SIZE);
1331
1332	memblock_dump_all();
1333}