Linux Audio

Check our new training course

Loading...
v5.4
   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
  20#include <asm/e820/api.h>
 
  21#include <asm/setup.h>
 
  22
  23/*
  24 * We organize the E820 table into three main data structures:
  25 *
  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:
  30 *
  31 *       - inform the user about the firmware's notion of memory layout
  32 *         via /sys/firmware/memmap
  33 *
  34 *       - the hibernation code uses it to generate a kernel-independent MD5
  35 *         fingerprint of the physical memory layout of a system.
  36 *
  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.
  43 *   We use this to:
  44 *
  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.
  49 *
  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.
  53 *
  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.
  58 */
  59static struct e820_table e820_table_init		__initdata;
  60static struct e820_table e820_table_kexec_init		__initdata;
  61static struct e820_table e820_table_firmware_init	__initdata;
  62
  63struct e820_table *e820_table __refdata			= &e820_table_init;
  64struct e820_table *e820_table_kexec __refdata		= &e820_table_kexec_init;
  65struct e820_table *e820_table_firmware __refdata	= &e820_table_firmware_init;
  66
  67/* For PCI or other memory-mapped resources */
  68unsigned long pci_mem_start = 0xaeedbabe;
  69#ifdef CONFIG_PCI
  70EXPORT_SYMBOL(pci_mem_start);
  71#endif
  72
  73/*
  74 * This function checks if any part of the range <start,end> is mapped
  75 * with type.
  76 */
  77static bool _e820__mapped_any(struct e820_table *table,
  78			      u64 start, u64 end, enum e820_type type)
  79{
  80	int i;
  81
  82	for (i = 0; i < table->nr_entries; i++) {
  83		struct e820_entry *entry = &table->entries[i];
  84
  85		if (type && entry->type != type)
  86			continue;
  87		if (entry->addr >= end || entry->addr + entry->size <= start)
  88			continue;
  89		return true;
  90	}
  91	return false;
  92}
  93
  94bool e820__mapped_raw_any(u64 start, u64 end, enum e820_type type)
  95{
  96	return _e820__mapped_any(e820_table_firmware, start, end, type);
  97}
  98EXPORT_SYMBOL_GPL(e820__mapped_raw_any);
  99
 100bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
 101{
 102	return _e820__mapped_any(e820_table, start, end, type);
 103}
 104EXPORT_SYMBOL_GPL(e820__mapped_any);
 105
 106/*
 107 * This function checks if the entire <start,end> range is mapped with 'type'.
 108 *
 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.
 111 */
 112static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
 113					     enum e820_type type)
 114{
 115	int i;
 116
 117	for (i = 0; i < e820_table->nr_entries; i++) {
 118		struct e820_entry *entry = &e820_table->entries[i];
 119
 120		if (type && entry->type != type)
 121			continue;
 122
 123		/* Is the region (part) in overlap with the current region? */
 124		if (entry->addr >= end || entry->addr + entry->size <= start)
 125			continue;
 126
 127		/*
 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
 130		 */
 131		if (entry->addr <= start)
 132			start = entry->addr + entry->size;
 133
 134		/*
 135		 * If 'start' is now at or beyond 'end', we're done, full
 136		 * coverage of the desired range exists:
 137		 */
 138		if (start >= end)
 139			return entry;
 140	}
 141
 142	return NULL;
 143}
 144
 145/*
 146 * This function checks if the entire range <start,end> is mapped with type.
 147 */
 148bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
 149{
 150	return __e820__mapped_all(start, end, type);
 151}
 152
 153/*
 154 * This function returns the type associated with the range <start,end>.
 155 */
 156int e820__get_entry_type(u64 start, u64 end)
 157{
 158	struct e820_entry *entry = __e820__mapped_all(start, end, 0);
 159
 160	return entry ? entry->type : -EINVAL;
 161}
 162
 163/*
 164 * Add a memory region to the kernel E820 map.
 165 */
 166static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
 
 167{
 168	int x = table->nr_entries;
 169
 170	if (x >= ARRAY_SIZE(table->entries)) {
 171		pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
 172		       start, start + size - 1);
 
 173		return;
 174	}
 175
 176	table->entries[x].addr = start;
 177	table->entries[x].size = size;
 178	table->entries[x].type = type;
 179	table->nr_entries++;
 180}
 181
 182void __init e820__range_add(u64 start, u64 size, enum e820_type type)
 183{
 184	__e820__range_add(e820_table, start, size, type);
 185}
 186
 187static void __init e820_print_type(enum e820_type type)
 188{
 189	switch (type) {
 190	case E820_TYPE_RAM:		/* Fall through: */
 191	case E820_TYPE_RESERVED_KERN:	pr_cont("usable");			break;
 192	case E820_TYPE_RESERVED:	pr_cont("reserved");			break;
 193	case E820_TYPE_ACPI:		pr_cont("ACPI data");			break;
 194	case E820_TYPE_NVS:		pr_cont("ACPI NVS");			break;
 195	case E820_TYPE_UNUSABLE:	pr_cont("unusable");			break;
 196	case E820_TYPE_PMEM:		/* Fall through: */
 197	case E820_TYPE_PRAM:		pr_cont("persistent (type %u)", type);	break;
 198	default:			pr_cont("type %u", type);		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 199	}
 200}
 201
 202void __init e820__print_table(char *who)
 203{
 204	int i;
 205
 206	for (i = 0; i < e820_table->nr_entries; i++) {
 207		pr_info("%s: [mem %#018Lx-%#018Lx] ",
 208			who,
 209			e820_table->entries[i].addr,
 210			e820_table->entries[i].addr + e820_table->entries[i].size - 1);
 211
 212		e820_print_type(e820_table->entries[i].type);
 213		pr_cont("\n");
 214	}
 215}
 216
 217/*
 218 * Sanitize an E820 map.
 219 *
 220 * Some E820 layouts include overlapping entries. The following
 221 * replaces the original E820 map with a new one, removing overlaps,
 222 * and resolving conflicting memory types in favor of highest
 223 * numbered type.
 224 *
 225 * The input parameter 'entries' points to an array of 'struct
 226 * e820_entry' which on entry has elements in the range [0, *nr_entries)
 227 * valid, and which has space for up to max_nr_entries entries.
 228 * On return, the resulting sanitized E820 map entries will be in
 229 * overwritten in the same location, starting at 'entries'.
 
 
 
 
 
 230 *
 231 * The integer pointed to by nr_entries must be valid on entry (the
 232 * current number of valid entries located at 'entries'). If the
 233 * sanitizing succeeds the *nr_entries will be updated with the new
 234 * number of valid entries (something no more than max_nr_entries).
 235 *
 236 * The return value from e820__update_table() is zero if it
 237 * successfully 'sanitized' the map entries passed in, and is -1
 238 * if it did nothing, which can happen if either of (1) it was
 239 * only passed one map entry, or (2) any of the input map entries
 240 * were invalid (start + size < start, meaning that the size was
 241 * so big the described memory range wrapped around through zero.)
 242 *
 243 *	Visually we're performing the following
 244 *	(1,2,3,4 = memory types)...
 245 *
 246 *	Sample memory map (w/overlaps):
 247 *	   ____22__________________
 248 *	   ______________________4_
 249 *	   ____1111________________
 250 *	   _44_____________________
 251 *	   11111111________________
 252 *	   ____________________33__
 253 *	   ___________44___________
 254 *	   __________33333_________
 255 *	   ______________22________
 256 *	   ___________________2222_
 257 *	   _________111111111______
 258 *	   _____________________11_
 259 *	   _________________4______
 260 *
 261 *	Sanitized equivalent (no overlap):
 262 *	   1_______________________
 263 *	   _44_____________________
 264 *	   ___1____________________
 265 *	   ____22__________________
 266 *	   ______11________________
 267 *	   _________1______________
 268 *	   __________3_____________
 269 *	   ___________44___________
 270 *	   _____________33_________
 271 *	   _______________2________
 272 *	   ________________1_______
 273 *	   _________________4______
 274 *	   ___________________2____
 275 *	   ____________________33__
 276 *	   ______________________4_
 277 */
 278struct change_member {
 279	/* Pointer to the original entry: */
 280	struct e820_entry	*entry;
 281	/* Address for this change point: */
 282	unsigned long long	addr;
 283};
 284
 285static struct change_member	change_point_list[2*E820_MAX_ENTRIES]	__initdata;
 286static struct change_member	*change_point[2*E820_MAX_ENTRIES]	__initdata;
 287static struct e820_entry	*overlap_list[E820_MAX_ENTRIES]		__initdata;
 288static struct e820_entry	new_entries[E820_MAX_ENTRIES]		__initdata;
 289
 290static int __init cpcompare(const void *a, const void *b)
 291{
 292	struct change_member * const *app = a, * const *bpp = b;
 293	const struct change_member *ap = *app, *bp = *bpp;
 294
 295	/*
 296	 * Inputs are pointers to two elements of change_point[].  If their
 297	 * addresses are not equal, their difference dominates.  If the addresses
 298	 * are equal, then consider one that represents the end of its region
 299	 * to be greater than one that does not.
 300	 */
 301	if (ap->addr != bp->addr)
 302		return ap->addr > bp->addr ? 1 : -1;
 303
 304	return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
 305}
 306
 307int __init e820__update_table(struct e820_table *table)
 
 308{
 309	struct e820_entry *entries = table->entries;
 310	u32 max_nr_entries = ARRAY_SIZE(table->entries);
 311	enum e820_type current_type, last_type;
 
 
 312	unsigned long long last_addr;
 313	u32 new_nr_entries, overlap_entries;
 314	u32 i, chg_idx, chg_nr;
 
 
 
 315
 316	/* If there's only one memory region, don't bother: */
 317	if (table->nr_entries < 2)
 318		return -1;
 319
 320	BUG_ON(table->nr_entries > max_nr_entries);
 
 321
 322	/* Bail out if we find any unreasonable addresses in the map: */
 323	for (i = 0; i < table->nr_entries; i++) {
 324		if (entries[i].addr + entries[i].size < entries[i].addr)
 325			return -1;
 326	}
 327
 328	/* Create pointers for initial change-point information (for sorting): */
 329	for (i = 0; i < 2 * table->nr_entries; i++)
 330		change_point[i] = &change_point_list[i];
 331
 332	/*
 333	 * Record all known change-points (starting and ending addresses),
 334	 * omitting empty memory regions:
 335	 */
 336	chg_idx = 0;
 337	for (i = 0; i < table->nr_entries; i++)	{
 338		if (entries[i].size != 0) {
 339			change_point[chg_idx]->addr	= entries[i].addr;
 340			change_point[chg_idx++]->entry	= &entries[i];
 341			change_point[chg_idx]->addr	= entries[i].addr + entries[i].size;
 342			change_point[chg_idx++]->entry	= &entries[i];
 343		}
 344	}
 345	chg_nr = chg_idx;
 346
 347	/* Sort change-point list by memory addresses (low -> high): */
 348	sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
 349
 350	/* Create a new memory map, removing overlaps: */
 351	overlap_entries = 0;	 /* Number of entries in the overlap table */
 352	new_nr_entries = 0;	 /* Index for creating new map entries */
 353	last_type = 0;		 /* Start with undefined memory type */
 354	last_addr = 0;		 /* Start with 0 as last starting address */
 355
 356	/* Loop through change-points, determining effect on the new map: */
 357	for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
 358		/* Keep track of all overlapping entries */
 359		if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
 360			/* Add map entry to overlap list (> 1 entry implies an overlap) */
 361			overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
 
 
 
 
 
 362		} else {
 363			/* Remove entry from list (order independent, so swap with last): */
 
 
 
 364			for (i = 0; i < overlap_entries; i++) {
 365				if (overlap_list[i] == change_point[chg_idx]->entry)
 366					overlap_list[i] = overlap_list[overlap_entries-1];
 
 
 367			}
 368			overlap_entries--;
 369		}
 370		/*
 371		 * If there are overlapping entries, decide which
 372		 * "type" to use (larger value takes precedence --
 373		 * 1=usable, 2,3,4,4+=unusable)
 374		 */
 375		current_type = 0;
 376		for (i = 0; i < overlap_entries; i++) {
 377			if (overlap_list[i]->type > current_type)
 378				current_type = overlap_list[i]->type;
 379		}
 380
 381		/* Continue building up new map based on this information: */
 382		if (current_type != last_type || current_type == E820_TYPE_PRAM) {
 
 383			if (last_type != 0)	 {
 384				new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
 385				/* Move forward only if the new size was non-zero: */
 386				if (new_entries[new_nr_entries].size != 0)
 387					/* No more space left for new entries? */
 388					if (++new_nr_entries >= max_nr_entries)
 
 
 
 
 
 
 
 389						break;
 390			}
 391			if (current_type != 0)	{
 392				new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
 393				new_entries[new_nr_entries].type = current_type;
 394				last_addr = change_point[chg_idx]->addr;
 
 395			}
 396			last_type = current_type;
 397		}
 398	}
 
 
 399
 400	/* Copy the new entries into the original location: */
 401	memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
 402	table->nr_entries = new_nr_entries;
 403
 404	return 0;
 405}
 406
 407static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
 408{
 409	struct boot_e820_entry *entry = entries;
 410
 411	while (nr_entries) {
 412		u64 start = entry->addr;
 413		u64 size = entry->size;
 414		u64 end = start + size - 1;
 415		u32 type = entry->type;
 416
 417		/* Ignore the entry on 64-bit overflow: */
 418		if (start > end && likely(size))
 419			return -1;
 420
 421		e820__range_add(start, size, type);
 422
 423		entry++;
 424		nr_entries--;
 425	}
 426	return 0;
 427}
 428
 429/*
 430 * Copy the BIOS E820 map into a safe place.
 431 *
 432 * Sanity-check it while we're at it..
 433 *
 434 * If we're lucky and live on a modern system, the setup code
 435 * will have given us a memory map that we can use to properly
 436 * set up memory.  If we aren't, we'll fake a memory map.
 437 */
 438static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
 439{
 440	/* Only one memory region (or negative)? Ignore it */
 441	if (nr_entries < 2)
 442		return -1;
 443
 444	return __append_e820_table(entries, nr_entries);
 445}
 446
 447static u64 __init
 448__e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
 
 449{
 450	u64 end;
 451	unsigned int i;
 452	u64 real_updated_size = 0;
 453
 454	BUG_ON(old_type == new_type);
 455
 456	if (size > (ULLONG_MAX - start))
 457		size = ULLONG_MAX - start;
 458
 459	end = start + size;
 460	printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
 
 461	e820_print_type(old_type);
 462	pr_cont(" ==> ");
 463	e820_print_type(new_type);
 464	pr_cont("\n");
 465
 466	for (i = 0; i < table->nr_entries; i++) {
 467		struct e820_entry *entry = &table->entries[i];
 468		u64 final_start, final_end;
 469		u64 entry_end;
 470
 471		if (entry->type != old_type)
 472			continue;
 473
 474		entry_end = entry->addr + entry->size;
 475
 476		/* Completely covered by new range? */
 477		if (entry->addr >= start && entry_end <= end) {
 478			entry->type = new_type;
 479			real_updated_size += entry->size;
 480			continue;
 481		}
 482
 483		/* New range is completely covered? */
 484		if (entry->addr < start && entry_end > end) {
 485			__e820__range_add(table, start, size, new_type);
 486			__e820__range_add(table, end, entry_end - end, entry->type);
 487			entry->size = start - entry->addr;
 488			real_updated_size += size;
 489			continue;
 490		}
 491
 492		/* Partially covered: */
 493		final_start = max(start, entry->addr);
 494		final_end = min(end, entry_end);
 495		if (final_start >= final_end)
 496			continue;
 497
 498		__e820__range_add(table, final_start, final_end - final_start, new_type);
 
 499
 500		real_updated_size += final_end - final_start;
 501
 502		/*
 503		 * Left range could be head or tail, so need to update
 504		 * its size first:
 505		 */
 506		entry->size -= final_end - final_start;
 507		if (entry->addr < final_start)
 508			continue;
 509
 510		entry->addr = final_end;
 511	}
 512	return real_updated_size;
 513}
 514
 515u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
 
 516{
 517	return __e820__range_update(e820_table, start, size, old_type, new_type);
 518}
 519
 520static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type  new_type)
 
 521{
 522	return __e820__range_update(e820_table_kexec, start, size, old_type, new_type);
 
 523}
 524
 525/* Remove a range of memory from the E820 table: */
 526u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
 
 527{
 528	int i;
 529	u64 end;
 530	u64 real_removed_size = 0;
 531
 532	if (size > (ULLONG_MAX - start))
 533		size = ULLONG_MAX - start;
 534
 535	end = start + size;
 536	printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
 537	if (check_type)
 
 538		e820_print_type(old_type);
 539	pr_cont("\n");
 540
 541	for (i = 0; i < e820_table->nr_entries; i++) {
 542		struct e820_entry *entry = &e820_table->entries[i];
 543		u64 final_start, final_end;
 544		u64 entry_end;
 545
 546		if (check_type && entry->type != old_type)
 547			continue;
 548
 549		entry_end = entry->addr + entry->size;
 550
 551		/* Completely covered? */
 552		if (entry->addr >= start && entry_end <= end) {
 553			real_removed_size += entry->size;
 554			memset(entry, 0, sizeof(*entry));
 555			continue;
 556		}
 557
 558		/* Is the new range completely covered? */
 559		if (entry->addr < start && entry_end > end) {
 560			e820__range_add(end, entry_end - end, entry->type);
 561			entry->size = start - entry->addr;
 562			real_removed_size += size;
 563			continue;
 564		}
 565
 566		/* Partially covered: */
 567		final_start = max(start, entry->addr);
 568		final_end = min(end, entry_end);
 569		if (final_start >= final_end)
 570			continue;
 571
 572		real_removed_size += final_end - final_start;
 573
 574		/*
 575		 * Left range could be head or tail, so need to update
 576		 * the size first:
 577		 */
 578		entry->size -= final_end - final_start;
 579		if (entry->addr < final_start)
 580			continue;
 581
 582		entry->addr = final_end;
 583	}
 584	return real_removed_size;
 585}
 586
 587void __init e820__update_table_print(void)
 588{
 589	if (e820__update_table(e820_table))
 590		return;
 591
 592	pr_info("modified physical RAM map:\n");
 593	e820__print_table("modified");
 594}
 595
 596static void __init e820__update_table_kexec(void)
 597{
 598	e820__update_table(e820_table_kexec);
 
 599}
 600
 601#define MAX_GAP_END 0x100000000ull
 602
 603/*
 604 * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
 605 */
 606static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
 
 607{
 608	unsigned long long last = MAX_GAP_END;
 609	int i = e820_table->nr_entries;
 610	int found = 0;
 611
 
 
 612	while (--i >= 0) {
 613		unsigned long long start = e820_table->entries[i].addr;
 614		unsigned long long end = start + e820_table->entries[i].size;
 
 
 
 615
 616		/*
 617		 * Since "last" is at most 4GB, we know we'll
 618		 * fit in 32 bits if this condition is true:
 619		 */
 620		if (last > end) {
 621			unsigned long gap = last - end;
 622
 623			if (gap >= *gapsize) {
 624				*gapsize = gap;
 625				*gapstart = end;
 626				found = 1;
 627			}
 628		}
 629		if (start < last)
 630			last = start;
 631	}
 632	return found;
 633}
 634
 635/*
 636 * Search for the biggest gap in the low 32 bits of the E820
 637 * memory space. We pass this space to the PCI subsystem, so
 638 * that it can assign MMIO resources for hotplug or
 639 * unconfigured devices in.
 640 *
 641 * Hopefully the BIOS let enough space left.
 642 */
 643__init void e820__setup_pci_gap(void)
 644{
 645	unsigned long gapstart, gapsize;
 646	int found;
 647
 
 648	gapsize = 0x400000;
 649	found  = e820_search_gap(&gapstart, &gapsize);
 650
 651	if (!found) {
 652#ifdef CONFIG_X86_64
 
 653		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
 654		pr_err("Cannot find an available gap in the 32-bit address range\n");
 655		pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
 656#else
 657		gapstart = 0x10000000;
 658#endif
 659	}
 
 660
 661	/*
 662	 * e820__reserve_resources_late() protects stolen RAM already:
 663	 */
 664	pci_mem_start = gapstart;
 665
 666	pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
 667		gapstart, gapstart + gapsize - 1);
 668}
 669
 670/*
 671 * Called late during init, in free_initmem().
 672 *
 673 * Initial e820_table and e820_table_kexec are largish __initdata arrays.
 674 *
 675 * Copy them to a (usually much smaller) dynamically allocated area that is
 676 * sized precisely after the number of e820 entries.
 677 *
 678 * This is done after we've performed all the fixes and tweaks to the tables.
 679 * All functions which modify them are __init functions, which won't exist
 680 * after free_initmem().
 681 */
 682__init void e820__reallocate_tables(void)
 683{
 684	struct e820_table *n;
 685	int size;
 686
 687	size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
 688	n = kmemdup(e820_table, size, GFP_KERNEL);
 689	BUG_ON(!n);
 690	e820_table = n;
 691
 692	size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
 693	n = kmemdup(e820_table_kexec, size, GFP_KERNEL);
 694	BUG_ON(!n);
 695	e820_table_kexec = n;
 696
 697	size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
 698	n = kmemdup(e820_table_firmware, size, GFP_KERNEL);
 699	BUG_ON(!n);
 700	e820_table_firmware = n;
 701}
 702
 703/*
 704 * Because of the small fixed size of struct boot_params, only the first
 705 * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
 706 * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
 707 * struct setup_data, which is parsed here.
 708 */
 709void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
 710{
 711	int entries;
 712	struct boot_e820_entry *extmap;
 713	struct setup_data *sdata;
 714
 715	sdata = early_memremap(phys_addr, data_len);
 716	entries = sdata->len / sizeof(*extmap);
 717	extmap = (struct boot_e820_entry *)(sdata->data);
 718
 719	__append_e820_table(extmap, entries);
 720	e820__update_table(e820_table);
 721
 722	memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
 723	memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
 724
 725	early_memunmap(sdata, data_len);
 726	pr_info("extended physical RAM map:\n");
 727	e820__print_table("extended");
 728}
 729
 730/*
 
 
 731 * Find the ranges of physical addresses that do not correspond to
 732 * E820 RAM areas and register the corresponding pages as 'nosave' for
 733 * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
 734 *
 735 * This function requires the E820 map to be sorted and without any
 736 * overlapping entries.
 737 */
 738void __init e820__register_nosave_regions(unsigned long limit_pfn)
 739{
 740	int i;
 741	unsigned long pfn = 0;
 742
 743	for (i = 0; i < e820_table->nr_entries; i++) {
 744		struct e820_entry *entry = &e820_table->entries[i];
 745
 746		if (pfn < PFN_UP(entry->addr))
 747			register_nosave_region(pfn, PFN_UP(entry->addr));
 748
 749		pfn = PFN_DOWN(entry->addr + entry->size);
 750
 751		if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
 752			register_nosave_region(PFN_UP(entry->addr), pfn);
 753
 754		if (pfn >= limit_pfn)
 755			break;
 756	}
 757}
 
 758
 759#ifdef CONFIG_ACPI
 760/*
 761 * Register ACPI NVS memory regions, so that we can save/restore them during
 762 * hibernation and the subsequent resume:
 763 */
 764static int __init e820__register_nvs_regions(void)
 765{
 766	int i;
 767
 768	for (i = 0; i < e820_table->nr_entries; i++) {
 769		struct e820_entry *entry = &e820_table->entries[i];
 770
 771		if (entry->type == E820_TYPE_NVS)
 772			acpi_nvs_register(entry->addr, entry->size);
 773	}
 774
 775	return 0;
 776}
 777core_initcall(e820__register_nvs_regions);
 778#endif
 779
 780/*
 781 * Allocate the requested number of bytes with the requsted alignment
 782 * and return (the physical address) to the caller. Also register this
 783 * range in the 'kexec' E820 table as a reserved range.
 784 *
 785 * This allows kexec to fake a new mptable, as if it came from the real
 786 * system.
 787 */
 788u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
 789{
 790	u64 addr;
 791
 792	addr = memblock_phys_alloc(size, align);
 793	if (addr) {
 794		e820__range_update_kexec(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
 795		pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
 796		e820__update_table_kexec();
 797	}
 798
 799	return addr;
 800}
 801
 802#ifdef CONFIG_X86_32
 803# ifdef CONFIG_X86_PAE
 804#  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
 805# else
 806#  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
 807# endif
 808#else /* CONFIG_X86_32 */
 809# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
 810#endif
 811
 812/*
 813 * Find the highest page frame number we have available
 814 */
 815static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
 816{
 817	int i;
 818	unsigned long last_pfn = 0;
 819	unsigned long max_arch_pfn = MAX_ARCH_PFN;
 820
 821	for (i = 0; i < e820_table->nr_entries; i++) {
 822		struct e820_entry *entry = &e820_table->entries[i];
 823		unsigned long start_pfn;
 824		unsigned long end_pfn;
 825
 826		if (entry->type != type)
 
 
 
 
 827			continue;
 828
 829		start_pfn = entry->addr >> PAGE_SHIFT;
 830		end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
 831
 832		if (start_pfn >= limit_pfn)
 833			continue;
 834		if (end_pfn > limit_pfn) {
 835			last_pfn = limit_pfn;
 836			break;
 837		}
 838		if (end_pfn > last_pfn)
 839			last_pfn = end_pfn;
 840	}
 841
 842	if (last_pfn > max_arch_pfn)
 843		last_pfn = max_arch_pfn;
 844
 845	pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
 846		last_pfn, max_arch_pfn);
 847	return last_pfn;
 848}
 849
 850unsigned long __init e820__end_of_ram_pfn(void)
 851{
 852	return e820_end_pfn(MAX_ARCH_PFN, E820_TYPE_RAM);
 853}
 854
 855unsigned long __init e820__end_of_low_ram_pfn(void)
 856{
 857	return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
 858}
 859
 860static void __init early_panic(char *msg)
 861{
 862	early_printk(msg);
 863	panic(msg);
 864}
 865
 866static int userdef __initdata;
 867
 868/* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
 869static int __init parse_memopt(char *p)
 870{
 871	u64 mem_size;
 872
 873	if (!p)
 874		return -EINVAL;
 875
 876	if (!strcmp(p, "nopentium")) {
 877#ifdef CONFIG_X86_32
 878		setup_clear_cpu_cap(X86_FEATURE_PSE);
 879		return 0;
 880#else
 881		pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
 882		return -EINVAL;
 883#endif
 884	}
 885
 886	userdef = 1;
 887	mem_size = memparse(p, &p);
 888
 889	/* Don't remove all memory when getting "mem={invalid}" parameter: */
 890	if (mem_size == 0)
 891		return -EINVAL;
 892
 893	e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
 894
 895#ifdef CONFIG_MEMORY_HOTPLUG
 896	max_mem_size = mem_size;
 897#endif
 898
 899	return 0;
 900}
 901early_param("mem", parse_memopt);
 902
 903static int __init parse_memmap_one(char *p)
 904{
 905	char *oldp;
 906	u64 start_at, mem_size;
 907
 908	if (!p)
 909		return -EINVAL;
 910
 911	if (!strncmp(p, "exactmap", 8)) {
 912#ifdef CONFIG_CRASH_DUMP
 913		/*
 914		 * If we are doing a crash dump, we still need to know
 915		 * the real memory size before the original memory map is
 916		 * reset.
 917		 */
 918		saved_max_pfn = e820__end_of_ram_pfn();
 919#endif
 920		e820_table->nr_entries = 0;
 921		userdef = 1;
 922		return 0;
 923	}
 924
 925	oldp = p;
 926	mem_size = memparse(p, &p);
 927	if (p == oldp)
 928		return -EINVAL;
 929
 930	userdef = 1;
 931	if (*p == '@') {
 932		start_at = memparse(p+1, &p);
 933		e820__range_add(start_at, mem_size, E820_TYPE_RAM);
 934	} else if (*p == '#') {
 935		start_at = memparse(p+1, &p);
 936		e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
 937	} else if (*p == '$') {
 938		start_at = memparse(p+1, &p);
 939		e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
 940	} else if (*p == '!') {
 941		start_at = memparse(p+1, &p);
 942		e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
 943	} else if (*p == '%') {
 944		enum e820_type from = 0, to = 0;
 945
 946		start_at = memparse(p + 1, &p);
 947		if (*p == '-')
 948			from = simple_strtoull(p + 1, &p, 0);
 949		if (*p == '+')
 950			to = simple_strtoull(p + 1, &p, 0);
 951		if (*p != '\0')
 952			return -EINVAL;
 953		if (from && to)
 954			e820__range_update(start_at, mem_size, from, to);
 955		else if (to)
 956			e820__range_add(start_at, mem_size, to);
 957		else if (from)
 958			e820__range_remove(start_at, mem_size, from, 1);
 959		else
 960			e820__range_remove(start_at, mem_size, 0, 0);
 961	} else {
 962		e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
 963	}
 964
 965	return *p == '\0' ? 0 : -EINVAL;
 966}
 967
 968static int __init parse_memmap_opt(char *str)
 969{
 970	while (str) {
 971		char *k = strchr(str, ',');
 972
 973		if (k)
 974			*k++ = 0;
 975
 976		parse_memmap_one(str);
 977		str = k;
 978	}
 979
 980	return 0;
 981}
 982early_param("memmap", parse_memmap_opt);
 983
 984/*
 985 * Reserve all entries from the bootloader's extensible data nodes list,
 986 * because if present we are going to use it later on to fetch e820
 987 * entries from it:
 988 */
 989void __init e820__reserve_setup_data(void)
 990{
 991	struct setup_data *data;
 992	u64 pa_data;
 993
 994	pa_data = boot_params.hdr.setup_data;
 995	if (!pa_data)
 996		return;
 997
 998	while (pa_data) {
 999		data = early_memremap(pa_data, sizeof(*data));
1000		e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1001		e820__range_update_kexec(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1002		pa_data = data->next;
1003		early_memunmap(data, sizeof(*data));
1004	}
1005
1006	e820__update_table(e820_table);
1007	e820__update_table(e820_table_kexec);
1008
1009	pr_info("extended physical RAM map:\n");
1010	e820__print_table("reserve setup_data");
1011}
1012
1013/*
1014 * Called after parse_early_param(), after early parameters (such as mem=)
1015 * have been processed, in which case we already have an E820 table filled in
1016 * via the parameter callback function(s), but it's not sorted and printed yet:
1017 */
1018void __init e820__finish_early_params(void)
1019{
1020	if (userdef) {
1021		if (e820__update_table(e820_table) < 0)
 
1022			early_panic("Invalid user supplied memory map");
1023
1024		pr_info("user-defined physical RAM map:\n");
1025		e820__print_table("user");
1026	}
1027}
1028
1029static const char *__init e820_type_to_string(struct e820_entry *entry)
1030{
1031	switch (entry->type) {
1032	case E820_TYPE_RESERVED_KERN:	/* Fall-through: */
1033	case E820_TYPE_RAM:		return "System RAM";
1034	case E820_TYPE_ACPI:		return "ACPI Tables";
1035	case E820_TYPE_NVS:		return "ACPI Non-volatile Storage";
1036	case E820_TYPE_UNUSABLE:	return "Unusable memory";
1037	case E820_TYPE_PRAM:		return "Persistent Memory (legacy)";
1038	case E820_TYPE_PMEM:		return "Persistent Memory";
1039	case E820_TYPE_RESERVED:	return "Reserved";
1040	default:			return "Unknown E820 type";
1041	}
1042}
1043
1044static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1045{
1046	switch (entry->type) {
1047	case E820_TYPE_RESERVED_KERN:	/* Fall-through: */
1048	case E820_TYPE_RAM:		return IORESOURCE_SYSTEM_RAM;
1049	case E820_TYPE_ACPI:		/* Fall-through: */
1050	case E820_TYPE_NVS:		/* Fall-through: */
1051	case E820_TYPE_UNUSABLE:	/* Fall-through: */
1052	case E820_TYPE_PRAM:		/* Fall-through: */
1053	case E820_TYPE_PMEM:		/* Fall-through: */
1054	case E820_TYPE_RESERVED:	/* Fall-through: */
1055	default:			return IORESOURCE_MEM;
 
1056	}
1057}
1058
1059static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1060{
1061	switch (entry->type) {
1062	case E820_TYPE_ACPI:		return IORES_DESC_ACPI_TABLES;
1063	case E820_TYPE_NVS:		return IORES_DESC_ACPI_NV_STORAGE;
1064	case E820_TYPE_PMEM:		return IORES_DESC_PERSISTENT_MEMORY;
1065	case E820_TYPE_PRAM:		return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1066	case E820_TYPE_RESERVED:	return IORES_DESC_RESERVED;
1067	case E820_TYPE_RESERVED_KERN:	/* Fall-through: */
1068	case E820_TYPE_RAM:		/* Fall-through: */
1069	case E820_TYPE_UNUSABLE:	/* Fall-through: */
1070	default:			return IORES_DESC_NONE;
 
 
 
 
1071	}
1072}
1073
1074static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1075{
1076	/* this is the legacy bios/dos rom-shadow + mmio region */
1077	if (res->start < (1ULL<<20))
1078		return true;
1079
1080	/*
1081	 * Treat persistent memory like device memory, i.e. reserve it
1082	 * for exclusive use of a driver
1083	 */
1084	switch (type) {
1085	case E820_TYPE_RESERVED:
1086	case E820_TYPE_PRAM:
1087	case E820_TYPE_PMEM:
1088		return false;
1089	case E820_TYPE_RESERVED_KERN:
1090	case E820_TYPE_RAM:
1091	case E820_TYPE_ACPI:
1092	case E820_TYPE_NVS:
1093	case E820_TYPE_UNUSABLE:
1094	default:
1095		return true;
1096	}
1097}
1098
1099/*
1100 * Mark E820 reserved areas as busy for the resource manager:
1101 */
1102
1103static struct resource __initdata *e820_res;
1104
1105void __init e820__reserve_resources(void)
1106{
1107	int i;
1108	struct resource *res;
1109	u64 end;
1110
1111	res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1112			     SMP_CACHE_BYTES);
1113	if (!res)
1114		panic("%s: Failed to allocate %zu bytes\n", __func__,
1115		      sizeof(*res) * e820_table->nr_entries);
1116	e820_res = res;
1117
1118	for (i = 0; i < e820_table->nr_entries; i++) {
1119		struct e820_entry *entry = e820_table->entries + i;
1120
1121		end = entry->addr + entry->size - 1;
1122		if (end != (resource_size_t)end) {
1123			res++;
1124			continue;
1125		}
1126		res->start = entry->addr;
1127		res->end   = end;
1128		res->name  = e820_type_to_string(entry);
1129		res->flags = e820_type_to_iomem_type(entry);
1130		res->desc  = e820_type_to_iores_desc(entry);
 
1131
1132		/*
1133		 * Don't register the region that could be conflicted with
1134		 * PCI device BAR resources and insert them later in
1135		 * pcibios_resource_survey():
1136		 */
1137		if (do_mark_busy(entry->type, res)) {
1138			res->flags |= IORESOURCE_BUSY;
1139			insert_resource(&iomem_resource, res);
1140		}
1141		res++;
1142	}
1143
1144	/* Expose the bootloader-provided memory layout to the sysfs. */
1145	for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1146		struct e820_entry *entry = e820_table_firmware->entries + i;
1147
1148		firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1149	}
1150}
1151
1152/*
1153 * How much should we pad the end of RAM, depending on where it is?
1154 */
1155static unsigned long __init ram_alignment(resource_size_t pos)
1156{
1157	unsigned long mb = pos >> 20;
1158
1159	/* To 64kB in the first megabyte */
1160	if (!mb)
1161		return 64*1024;
1162
1163	/* To 1MB in the first 16MB */
1164	if (mb < 16)
1165		return 1024*1024;
1166
1167	/* To 64MB for anything above that */
1168	return 64*1024*1024;
1169}
1170
1171#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1172
1173void __init e820__reserve_resources_late(void)
1174{
1175	int i;
1176	struct resource *res;
1177
1178	res = e820_res;
1179	for (i = 0; i < e820_table->nr_entries; i++) {
1180		if (!res->parent && res->end)
1181			insert_resource_expand_to_fit(&iomem_resource, res);
1182		res++;
1183	}
1184
1185	/*
1186	 * Try to bump up RAM regions to reasonable boundaries, to
1187	 * avoid stolen RAM:
1188	 */
1189	for (i = 0; i < e820_table->nr_entries; i++) {
1190		struct e820_entry *entry = &e820_table->entries[i];
1191		u64 start, end;
1192
1193		if (entry->type != E820_TYPE_RAM)
1194			continue;
1195
1196		start = entry->addr + entry->size;
1197		end = round_up(start, ram_alignment(start)) - 1;
1198		if (end > MAX_RESOURCE_SIZE)
1199			end = MAX_RESOURCE_SIZE;
1200		if (start >= end)
1201			continue;
1202
1203		printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1204		reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
 
 
1205	}
1206}
1207
1208/*
1209 * Pass the firmware (bootloader) E820 map to the kernel and process it:
1210 */
1211char *__init e820__memory_setup_default(void)
1212{
1213	char *who = "BIOS-e820";
1214
1215	/*
1216	 * Try to copy the BIOS-supplied E820-map.
1217	 *
1218	 * Otherwise fake a memory map; one section from 0k->640k,
1219	 * the next section from 1mb->appropriate_mem_k
1220	 */
1221	if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
 
 
 
 
 
 
1222		u64 mem_size;
1223
1224		/* Compare results from other methods and take the one that gives more RAM: */
1225		if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
 
1226			mem_size = boot_params.screen_info.ext_mem_k;
1227			who = "BIOS-88";
1228		} else {
1229			mem_size = boot_params.alt_mem_k;
1230			who = "BIOS-e801";
1231		}
1232
1233		e820_table->nr_entries = 0;
1234		e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1235		e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1236	}
1237
1238	/* We just appended a lot of ranges, sanitize the table: */
1239	e820__update_table(e820_table);
1240
1241	return who;
1242}
1243
1244/*
1245 * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1246 * E820 map - with an optional platform quirk available for virtual platforms
1247 * to override this method of boot environment processing:
1248 */
1249void __init e820__memory_setup(void)
1250{
1251	char *who;
1252
1253	/* This is a firmware interface ABI - make sure we don't break it: */
1254	BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1255
1256	who = x86_init.resources.memory_setup();
1257
1258	memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1259	memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1260
1261	pr_info("BIOS-provided physical RAM map:\n");
1262	e820__print_table(who);
1263}
1264
1265void __init e820__memblock_setup(void)
1266{
1267	int i;
1268	u64 end;
1269
1270	/*
1271	 * The bootstrap memblock region count maximum is 128 entries
1272	 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1273	 * than that - so allow memblock resizing.
1274	 *
1275	 * This is safe, because this call happens pretty late during x86 setup,
1276	 * so we know about reserved memory regions already. (This is important
1277	 * so that memblock resizing does no stomp over reserved areas.)
1278	 */
1279	memblock_allow_resize();
1280
1281	for (i = 0; i < e820_table->nr_entries; i++) {
1282		struct e820_entry *entry = &e820_table->entries[i];
1283
1284		end = entry->addr + entry->size;
1285		if (end != (resource_size_t)end)
1286			continue;
1287
1288		if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1289			continue;
1290
1291		memblock_add(entry->addr, entry->size);
1292	}
1293
1294	/* Throw away partial pages: */
1295	memblock_trim_memory(PAGE_SIZE);
1296
1297	memblock_dump_all();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1298}
v4.6
 
   1/*
   2 * Handle the memory map.
   3 * The functions here do the job until bootmem takes over.
   4 *
   5 *  Getting sanitize_e820_map() in sync with i386 version by applying change:
   6 *  -  Provisions for empty E820 memory regions (reported by certain BIOSes).
   7 *     Alex Achenbach <xela@slit.de>, December 2002.
   8 *  Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
   9 *
 
 
 
  10 */
  11#include <linux/kernel.h>
  12#include <linux/types.h>
  13#include <linux/init.h>
  14#include <linux/crash_dump.h>
  15#include <linux/export.h>
  16#include <linux/bootmem.h>
  17#include <linux/pfn.h>
  18#include <linux/suspend.h>
  19#include <linux/acpi.h>
  20#include <linux/firmware-map.h>
  21#include <linux/memblock.h>
  22#include <linux/sort.h>
 
  23
  24#include <asm/e820.h>
  25#include <asm/proto.h>
  26#include <asm/setup.h>
  27#include <asm/cpufeature.h>
  28
  29/*
  30 * The e820 map is the map that gets modified e.g. with command line parameters
  31 * and that is also registered with modifications in the kernel resource tree
  32 * with the iomem_resource as parent.
 
 
 
 
 
 
 
 
 
  33 *
  34 * The e820_saved is directly saved after the BIOS-provided memory map is
  35 * copied. It doesn't get modified afterwards. It's registered for the
  36 * /sys/firmware/memmap interface.
 
 
 
 
  37 *
  38 * That memory map is not modified and is used as base for kexec. The kexec'd
  39 * kernel should get the same memory map as the firmware provides. Then the
  40 * user can e.g. boot the original kernel with mem=1G while still booting the
  41 * next kernel with full memory.
 
 
 
 
 
 
 
 
 
  42 */
  43struct e820map e820;
  44struct e820map e820_saved;
 
 
 
 
 
  45
  46/* For PCI or other memory-mapped resources */
  47unsigned long pci_mem_start = 0xaeedbabe;
  48#ifdef CONFIG_PCI
  49EXPORT_SYMBOL(pci_mem_start);
  50#endif
  51
  52/*
  53 * This function checks if any part of the range <start,end> is mapped
  54 * with type.
  55 */
  56int
  57e820_any_mapped(u64 start, u64 end, unsigned type)
  58{
  59	int i;
  60
  61	for (i = 0; i < e820.nr_map; i++) {
  62		struct e820entry *ei = &e820.map[i];
  63
  64		if (type && ei->type != type)
  65			continue;
  66		if (ei->addr >= end || ei->addr + ei->size <= start)
  67			continue;
  68		return 1;
  69	}
  70	return 0;
 
 
 
 
 
 
 
 
 
 
 
  71}
  72EXPORT_SYMBOL_GPL(e820_any_mapped);
  73
  74/*
  75 * This function checks if the entire range <start,end> is mapped with type.
  76 *
  77 * Note: this function only works correct if the e820 table is sorted and
  78 * not-overlapping, which is the case
  79 */
  80int __init e820_all_mapped(u64 start, u64 end, unsigned type)
 
  81{
  82	int i;
  83
  84	for (i = 0; i < e820.nr_map; i++) {
  85		struct e820entry *ei = &e820.map[i];
  86
  87		if (type && ei->type != type)
  88			continue;
  89		/* is the region (part) in overlap with the current region ?*/
  90		if (ei->addr >= end || ei->addr + ei->size <= start)
 
  91			continue;
  92
  93		/* if the region is at the beginning of <start,end> we move
  94		 * start to the end of the region since it's ok until there
 
  95		 */
  96		if (ei->addr <= start)
  97			start = ei->addr + ei->size;
 
  98		/*
  99		 * if start is now at or beyond end, we're done, full
 100		 * coverage
 101		 */
 102		if (start >= end)
 103			return 1;
 104	}
 105	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 106}
 107
 108/*
 109 * Add a memory region to the kernel e820 map.
 110 */
 111static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
 112					 int type)
 113{
 114	int x = e820x->nr_map;
 115
 116	if (x >= ARRAY_SIZE(e820x->map)) {
 117		printk(KERN_ERR "e820: too many entries; ignoring [mem %#010llx-%#010llx]\n",
 118		       (unsigned long long) start,
 119		       (unsigned long long) (start + size - 1));
 120		return;
 121	}
 122
 123	e820x->map[x].addr = start;
 124	e820x->map[x].size = size;
 125	e820x->map[x].type = type;
 126	e820x->nr_map++;
 127}
 128
 129void __init e820_add_region(u64 start, u64 size, int type)
 130{
 131	__e820_add_region(&e820, start, size, type);
 132}
 133
 134static void __init e820_print_type(u32 type)
 135{
 136	switch (type) {
 137	case E820_RAM:
 138	case E820_RESERVED_KERN:
 139		printk(KERN_CONT "usable");
 140		break;
 141	case E820_RESERVED:
 142		printk(KERN_CONT "reserved");
 143		break;
 144	case E820_ACPI:
 145		printk(KERN_CONT "ACPI data");
 146		break;
 147	case E820_NVS:
 148		printk(KERN_CONT "ACPI NVS");
 149		break;
 150	case E820_UNUSABLE:
 151		printk(KERN_CONT "unusable");
 152		break;
 153	case E820_PMEM:
 154	case E820_PRAM:
 155		printk(KERN_CONT "persistent (type %u)", type);
 156		break;
 157	default:
 158		printk(KERN_CONT "type %u", type);
 159		break;
 160	}
 161}
 162
 163void __init e820_print_map(char *who)
 164{
 165	int i;
 166
 167	for (i = 0; i < e820.nr_map; i++) {
 168		printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who,
 169		       (unsigned long long) e820.map[i].addr,
 170		       (unsigned long long)
 171		       (e820.map[i].addr + e820.map[i].size - 1));
 172		e820_print_type(e820.map[i].type);
 173		printk(KERN_CONT "\n");
 
 174	}
 175}
 176
 177/*
 178 * Sanitize the BIOS e820 map.
 179 *
 180 * Some e820 responses include overlapping entries. The following
 181 * replaces the original e820 map with a new one, removing overlaps,
 182 * and resolving conflicting memory types in favor of highest
 183 * numbered type.
 184 *
 185 * The input parameter biosmap points to an array of 'struct
 186 * e820entry' which on entry has elements in the range [0, *pnr_map)
 187 * valid, and which has space for up to max_nr_map entries.
 188 * On return, the resulting sanitized e820 map entries will be in
 189 * overwritten in the same location, starting at biosmap.
 190 *
 191 * The integer pointed to by pnr_map must be valid on entry (the
 192 * current number of valid entries located at biosmap). If the
 193 * sanitizing succeeds the *pnr_map will be updated with the new
 194 * number of valid entries (something no more than max_nr_map).
 195 *
 196 * The return value from sanitize_e820_map() is zero if it
 
 
 
 
 
 197 * successfully 'sanitized' the map entries passed in, and is -1
 198 * if it did nothing, which can happen if either of (1) it was
 199 * only passed one map entry, or (2) any of the input map entries
 200 * were invalid (start + size < start, meaning that the size was
 201 * so big the described memory range wrapped around through zero.)
 202 *
 203 *	Visually we're performing the following
 204 *	(1,2,3,4 = memory types)...
 205 *
 206 *	Sample memory map (w/overlaps):
 207 *	   ____22__________________
 208 *	   ______________________4_
 209 *	   ____1111________________
 210 *	   _44_____________________
 211 *	   11111111________________
 212 *	   ____________________33__
 213 *	   ___________44___________
 214 *	   __________33333_________
 215 *	   ______________22________
 216 *	   ___________________2222_
 217 *	   _________111111111______
 218 *	   _____________________11_
 219 *	   _________________4______
 220 *
 221 *	Sanitized equivalent (no overlap):
 222 *	   1_______________________
 223 *	   _44_____________________
 224 *	   ___1____________________
 225 *	   ____22__________________
 226 *	   ______11________________
 227 *	   _________1______________
 228 *	   __________3_____________
 229 *	   ___________44___________
 230 *	   _____________33_________
 231 *	   _______________2________
 232 *	   ________________1_______
 233 *	   _________________4______
 234 *	   ___________________2____
 235 *	   ____________________33__
 236 *	   ______________________4_
 237 */
 238struct change_member {
 239	struct e820entry *pbios; /* pointer to original bios entry */
 240	unsigned long long addr; /* address for this change point */
 
 
 241};
 242
 
 
 
 
 
 243static int __init cpcompare(const void *a, const void *b)
 244{
 245	struct change_member * const *app = a, * const *bpp = b;
 246	const struct change_member *ap = *app, *bp = *bpp;
 247
 248	/*
 249	 * Inputs are pointers to two elements of change_point[].  If their
 250	 * addresses are unequal, their difference dominates.  If the addresses
 251	 * are equal, then consider one that represents the end of its region
 252	 * to be greater than one that does not.
 253	 */
 254	if (ap->addr != bp->addr)
 255		return ap->addr > bp->addr ? 1 : -1;
 256
 257	return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr);
 258}
 259
 260int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
 261			     u32 *pnr_map)
 262{
 263	static struct change_member change_point_list[2*E820_X_MAX] __initdata;
 264	static struct change_member *change_point[2*E820_X_MAX] __initdata;
 265	static struct e820entry *overlap_list[E820_X_MAX] __initdata;
 266	static struct e820entry new_bios[E820_X_MAX] __initdata;
 267	unsigned long current_type, last_type;
 268	unsigned long long last_addr;
 269	int chgidx;
 270	int overlap_entries;
 271	int new_bios_entry;
 272	int old_nr, new_nr, chg_nr;
 273	int i;
 274
 275	/* if there's only one memory region, don't bother */
 276	if (*pnr_map < 2)
 277		return -1;
 278
 279	old_nr = *pnr_map;
 280	BUG_ON(old_nr > max_nr_map);
 281
 282	/* bail out if we find any unreasonable addresses in bios map */
 283	for (i = 0; i < old_nr; i++)
 284		if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 285			return -1;
 
 286
 287	/* create pointers for initial change-point information (for sorting) */
 288	for (i = 0; i < 2 * old_nr; i++)
 289		change_point[i] = &change_point_list[i];
 290
 291	/* record all known change-points (starting and ending addresses),
 292	   omitting those that are for empty memory regions */
 293	chgidx = 0;
 294	for (i = 0; i < old_nr; i++)	{
 295		if (biosmap[i].size != 0) {
 296			change_point[chgidx]->addr = biosmap[i].addr;
 297			change_point[chgidx++]->pbios = &biosmap[i];
 298			change_point[chgidx]->addr = biosmap[i].addr +
 299				biosmap[i].size;
 300			change_point[chgidx++]->pbios = &biosmap[i];
 
 301		}
 302	}
 303	chg_nr = chgidx;
 304
 305	/* sort change-point list by memory addresses (low -> high) */
 306	sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL);
 307
 308	/* create a new bios memory map, removing overlaps */
 309	overlap_entries = 0;	 /* number of entries in the overlap table */
 310	new_bios_entry = 0;	 /* index for creating new bios map entries */
 311	last_type = 0;		 /* start with undefined memory type */
 312	last_addr = 0;		 /* start with 0 as last starting address */
 313
 314	/* loop through change-points, determining affect on the new bios map */
 315	for (chgidx = 0; chgidx < chg_nr; chgidx++) {
 316		/* keep track of all overlapping bios entries */
 317		if (change_point[chgidx]->addr ==
 318		    change_point[chgidx]->pbios->addr) {
 319			/*
 320			 * add map entry to overlap list (> 1 entry
 321			 * implies an overlap)
 322			 */
 323			overlap_list[overlap_entries++] =
 324				change_point[chgidx]->pbios;
 325		} else {
 326			/*
 327			 * remove entry from list (order independent,
 328			 * so swap with last)
 329			 */
 330			for (i = 0; i < overlap_entries; i++) {
 331				if (overlap_list[i] ==
 332				    change_point[chgidx]->pbios)
 333					overlap_list[i] =
 334						overlap_list[overlap_entries-1];
 335			}
 336			overlap_entries--;
 337		}
 338		/*
 339		 * if there are overlapping entries, decide which
 340		 * "type" to use (larger value takes precedence --
 341		 * 1=usable, 2,3,4,4+=unusable)
 342		 */
 343		current_type = 0;
 344		for (i = 0; i < overlap_entries; i++)
 345			if (overlap_list[i]->type > current_type)
 346				current_type = overlap_list[i]->type;
 347		/*
 348		 * continue building up new bios map based on this
 349		 * information
 350		 */
 351		if (current_type != last_type || current_type == E820_PRAM) {
 352			if (last_type != 0)	 {
 353				new_bios[new_bios_entry].size =
 354					change_point[chgidx]->addr - last_addr;
 355				/*
 356				 * move forward only if the new size
 357				 * was non-zero
 358				 */
 359				if (new_bios[new_bios_entry].size != 0)
 360					/*
 361					 * no more space left for new
 362					 * bios entries ?
 363					 */
 364					if (++new_bios_entry >= max_nr_map)
 365						break;
 366			}
 367			if (current_type != 0)	{
 368				new_bios[new_bios_entry].addr =
 369					change_point[chgidx]->addr;
 370				new_bios[new_bios_entry].type = current_type;
 371				last_addr = change_point[chgidx]->addr;
 372			}
 373			last_type = current_type;
 374		}
 375	}
 376	/* retain count for new bios entries */
 377	new_nr = new_bios_entry;
 378
 379	/* copy new bios mapping into original location */
 380	memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
 381	*pnr_map = new_nr;
 382
 383	return 0;
 384}
 385
 386static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
 387{
 388	while (nr_map) {
 389		u64 start = biosmap->addr;
 390		u64 size = biosmap->size;
 391		u64 end = start + size;
 392		u32 type = biosmap->type;
 
 
 393
 394		/* Overflow in 64 bits? Ignore the memory map. */
 395		if (start > end)
 396			return -1;
 397
 398		e820_add_region(start, size, type);
 399
 400		biosmap++;
 401		nr_map--;
 402	}
 403	return 0;
 404}
 405
 406/*
 407 * Copy the BIOS e820 map into a safe place.
 408 *
 409 * Sanity-check it while we're at it..
 410 *
 411 * If we're lucky and live on a modern system, the setup code
 412 * will have given us a memory map that we can use to properly
 413 * set up memory.  If we aren't, we'll fake a memory map.
 414 */
 415static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
 416{
 417	/* Only one memory region (or negative)? Ignore it */
 418	if (nr_map < 2)
 419		return -1;
 420
 421	return __append_e820_map(biosmap, nr_map);
 422}
 423
 424static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
 425					u64 size, unsigned old_type,
 426					unsigned new_type)
 427{
 428	u64 end;
 429	unsigned int i;
 430	u64 real_updated_size = 0;
 431
 432	BUG_ON(old_type == new_type);
 433
 434	if (size > (ULLONG_MAX - start))
 435		size = ULLONG_MAX - start;
 436
 437	end = start + size;
 438	printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ",
 439	       (unsigned long long) start, (unsigned long long) (end - 1));
 440	e820_print_type(old_type);
 441	printk(KERN_CONT " ==> ");
 442	e820_print_type(new_type);
 443	printk(KERN_CONT "\n");
 444
 445	for (i = 0; i < e820x->nr_map; i++) {
 446		struct e820entry *ei = &e820x->map[i];
 447		u64 final_start, final_end;
 448		u64 ei_end;
 449
 450		if (ei->type != old_type)
 451			continue;
 452
 453		ei_end = ei->addr + ei->size;
 454		/* totally covered by new range? */
 455		if (ei->addr >= start && ei_end <= end) {
 456			ei->type = new_type;
 457			real_updated_size += ei->size;
 
 458			continue;
 459		}
 460
 461		/* new range is totally covered? */
 462		if (ei->addr < start && ei_end > end) {
 463			__e820_add_region(e820x, start, size, new_type);
 464			__e820_add_region(e820x, end, ei_end - end, ei->type);
 465			ei->size = start - ei->addr;
 466			real_updated_size += size;
 467			continue;
 468		}
 469
 470		/* partially covered */
 471		final_start = max(start, ei->addr);
 472		final_end = min(end, ei_end);
 473		if (final_start >= final_end)
 474			continue;
 475
 476		__e820_add_region(e820x, final_start, final_end - final_start,
 477				  new_type);
 478
 479		real_updated_size += final_end - final_start;
 480
 481		/*
 482		 * left range could be head or tail, so need to update
 483		 * size at first.
 484		 */
 485		ei->size -= final_end - final_start;
 486		if (ei->addr < final_start)
 487			continue;
 488		ei->addr = final_end;
 
 489	}
 490	return real_updated_size;
 491}
 492
 493u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
 494			     unsigned new_type)
 495{
 496	return __e820_update_range(&e820, start, size, old_type, new_type);
 497}
 498
 499static u64 __init e820_update_range_saved(u64 start, u64 size,
 500					  unsigned old_type, unsigned new_type)
 501{
 502	return __e820_update_range(&e820_saved, start, size, old_type,
 503				     new_type);
 504}
 505
 506/* make e820 not cover the range */
 507u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
 508			     int checktype)
 509{
 510	int i;
 511	u64 end;
 512	u64 real_removed_size = 0;
 513
 514	if (size > (ULLONG_MAX - start))
 515		size = ULLONG_MAX - start;
 516
 517	end = start + size;
 518	printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ",
 519	       (unsigned long long) start, (unsigned long long) (end - 1));
 520	if (checktype)
 521		e820_print_type(old_type);
 522	printk(KERN_CONT "\n");
 523
 524	for (i = 0; i < e820.nr_map; i++) {
 525		struct e820entry *ei = &e820.map[i];
 526		u64 final_start, final_end;
 527		u64 ei_end;
 528
 529		if (checktype && ei->type != old_type)
 530			continue;
 531
 532		ei_end = ei->addr + ei->size;
 533		/* totally covered? */
 534		if (ei->addr >= start && ei_end <= end) {
 535			real_removed_size += ei->size;
 536			memset(ei, 0, sizeof(struct e820entry));
 
 537			continue;
 538		}
 539
 540		/* new range is totally covered? */
 541		if (ei->addr < start && ei_end > end) {
 542			e820_add_region(end, ei_end - end, ei->type);
 543			ei->size = start - ei->addr;
 544			real_removed_size += size;
 545			continue;
 546		}
 547
 548		/* partially covered */
 549		final_start = max(start, ei->addr);
 550		final_end = min(end, ei_end);
 551		if (final_start >= final_end)
 552			continue;
 
 553		real_removed_size += final_end - final_start;
 554
 555		/*
 556		 * left range could be head or tail, so need to update
 557		 * size at first.
 558		 */
 559		ei->size -= final_end - final_start;
 560		if (ei->addr < final_start)
 561			continue;
 562		ei->addr = final_end;
 
 563	}
 564	return real_removed_size;
 565}
 566
 567void __init update_e820(void)
 568{
 569	if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map))
 570		return;
 571	printk(KERN_INFO "e820: modified physical RAM map:\n");
 572	e820_print_map("modified");
 
 573}
 574static void __init update_e820_saved(void)
 
 575{
 576	sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map),
 577				&e820_saved.nr_map);
 578}
 
 579#define MAX_GAP_END 0x100000000ull
 
 580/*
 581 * Search for a gap in the e820 memory space from start_addr to end_addr.
 582 */
 583__init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
 584		unsigned long start_addr, unsigned long long end_addr)
 585{
 586	unsigned long long last;
 587	int i = e820.nr_map;
 588	int found = 0;
 589
 590	last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
 591
 592	while (--i >= 0) {
 593		unsigned long long start = e820.map[i].addr;
 594		unsigned long long end = start + e820.map[i].size;
 595
 596		if (end < start_addr)
 597			continue;
 598
 599		/*
 600		 * Since "last" is at most 4GB, we know we'll
 601		 * fit in 32 bits if this condition is true
 602		 */
 603		if (last > end) {
 604			unsigned long gap = last - end;
 605
 606			if (gap >= *gapsize) {
 607				*gapsize = gap;
 608				*gapstart = end;
 609				found = 1;
 610			}
 611		}
 612		if (start < last)
 613			last = start;
 614	}
 615	return found;
 616}
 617
 618/*
 619 * Search for the biggest gap in the low 32 bits of the e820
 620 * memory space.  We pass this space to PCI to assign MMIO resources
 621 * for hotplug or unconfigured devices in.
 
 
 622 * Hopefully the BIOS let enough space left.
 623 */
 624__init void e820_setup_gap(void)
 625{
 626	unsigned long gapstart, gapsize;
 627	int found;
 628
 629	gapstart = 0x10000000;
 630	gapsize = 0x400000;
 631	found  = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
 632
 
 633#ifdef CONFIG_X86_64
 634	if (!found) {
 635		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
 636		printk(KERN_ERR
 637	"e820: cannot find a gap in the 32bit address range\n"
 638	"e820: PCI devices with unassigned 32bit BARs may break!\n");
 
 
 639	}
 640#endif
 641
 642	/*
 643	 * e820_reserve_resources_late protect stolen RAM already
 644	 */
 645	pci_mem_start = gapstart;
 646
 647	printk(KERN_INFO
 648	       "e820: [mem %#010lx-%#010lx] available for PCI devices\n",
 649	       gapstart, gapstart + gapsize - 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 650}
 651
 652/**
 653 * Because of the size limitation of struct boot_params, only first
 654 * 128 E820 memory entries are passed to kernel via
 655 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
 656 * linked list of struct setup_data, which is parsed here.
 657 */
 658void __init parse_e820_ext(u64 phys_addr, u32 data_len)
 659{
 660	int entries;
 661	struct e820entry *extmap;
 662	struct setup_data *sdata;
 663
 664	sdata = early_memremap(phys_addr, data_len);
 665	entries = sdata->len / sizeof(struct e820entry);
 666	extmap = (struct e820entry *)(sdata->data);
 667	__append_e820_map(extmap, entries);
 668	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 
 
 
 
 
 669	early_memunmap(sdata, data_len);
 670	printk(KERN_INFO "e820: extended physical RAM map:\n");
 671	e820_print_map("extended");
 672}
 673
 674#if defined(CONFIG_X86_64) || \
 675	(defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
 676/**
 677 * Find the ranges of physical addresses that do not correspond to
 678 * e820 RAM areas and mark the corresponding pages as nosave for
 679 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
 680 *
 681 * This function requires the e820 map to be sorted and without any
 682 * overlapping entries.
 683 */
 684void __init e820_mark_nosave_regions(unsigned long limit_pfn)
 685{
 686	int i;
 687	unsigned long pfn = 0;
 688
 689	for (i = 0; i < e820.nr_map; i++) {
 690		struct e820entry *ei = &e820.map[i];
 691
 692		if (pfn < PFN_UP(ei->addr))
 693			register_nosave_region(pfn, PFN_UP(ei->addr));
 694
 695		pfn = PFN_DOWN(ei->addr + ei->size);
 696
 697		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
 698			register_nosave_region(PFN_UP(ei->addr), pfn);
 699
 700		if (pfn >= limit_pfn)
 701			break;
 702	}
 703}
 704#endif
 705
 706#ifdef CONFIG_ACPI
 707/**
 708 * Mark ACPI NVS memory region, so that we can save/restore it during
 709 * hibernation and the subsequent resume.
 710 */
 711static int __init e820_mark_nvs_memory(void)
 712{
 713	int i;
 714
 715	for (i = 0; i < e820.nr_map; i++) {
 716		struct e820entry *ei = &e820.map[i];
 717
 718		if (ei->type == E820_NVS)
 719			acpi_nvs_register(ei->addr, ei->size);
 720	}
 721
 722	return 0;
 723}
 724core_initcall(e820_mark_nvs_memory);
 725#endif
 726
 727/*
 728 * pre allocated 4k and reserved it in memblock and e820_saved
 
 
 
 
 
 729 */
 730u64 __init early_reserve_e820(u64 size, u64 align)
 731{
 732	u64 addr;
 733
 734	addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
 735	if (addr) {
 736		e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED);
 737		printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n");
 738		update_e820_saved();
 739	}
 740
 741	return addr;
 742}
 743
 744#ifdef CONFIG_X86_32
 745# ifdef CONFIG_X86_PAE
 746#  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
 747# else
 748#  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
 749# endif
 750#else /* CONFIG_X86_32 */
 751# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
 752#endif
 753
 754/*
 755 * Find the highest page frame number we have available
 756 */
 757static unsigned long __init e820_end_pfn(unsigned long limit_pfn)
 758{
 759	int i;
 760	unsigned long last_pfn = 0;
 761	unsigned long max_arch_pfn = MAX_ARCH_PFN;
 762
 763	for (i = 0; i < e820.nr_map; i++) {
 764		struct e820entry *ei = &e820.map[i];
 765		unsigned long start_pfn;
 766		unsigned long end_pfn;
 767
 768		/*
 769		 * Persistent memory is accounted as ram for purposes of
 770		 * establishing max_pfn and mem_map.
 771		 */
 772		if (ei->type != E820_RAM && ei->type != E820_PRAM)
 773			continue;
 774
 775		start_pfn = ei->addr >> PAGE_SHIFT;
 776		end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
 777
 778		if (start_pfn >= limit_pfn)
 779			continue;
 780		if (end_pfn > limit_pfn) {
 781			last_pfn = limit_pfn;
 782			break;
 783		}
 784		if (end_pfn > last_pfn)
 785			last_pfn = end_pfn;
 786	}
 787
 788	if (last_pfn > max_arch_pfn)
 789		last_pfn = max_arch_pfn;
 790
 791	printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
 792			 last_pfn, max_arch_pfn);
 793	return last_pfn;
 794}
 795unsigned long __init e820_end_of_ram_pfn(void)
 
 796{
 797	return e820_end_pfn(MAX_ARCH_PFN);
 798}
 799
 800unsigned long __init e820_end_of_low_ram_pfn(void)
 801{
 802	return e820_end_pfn(1UL << (32-PAGE_SHIFT));
 803}
 804
 805static void early_panic(char *msg)
 806{
 807	early_printk(msg);
 808	panic(msg);
 809}
 810
 811static int userdef __initdata;
 812
 813/* "mem=nopentium" disables the 4MB page tables. */
 814static int __init parse_memopt(char *p)
 815{
 816	u64 mem_size;
 817
 818	if (!p)
 819		return -EINVAL;
 820
 821	if (!strcmp(p, "nopentium")) {
 822#ifdef CONFIG_X86_32
 823		setup_clear_cpu_cap(X86_FEATURE_PSE);
 824		return 0;
 825#else
 826		printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
 827		return -EINVAL;
 828#endif
 829	}
 830
 831	userdef = 1;
 832	mem_size = memparse(p, &p);
 833	/* don't remove all of memory when handling "mem={invalid}" param */
 
 834	if (mem_size == 0)
 835		return -EINVAL;
 836	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
 
 
 
 
 
 837
 838	return 0;
 839}
 840early_param("mem", parse_memopt);
 841
 842static int __init parse_memmap_one(char *p)
 843{
 844	char *oldp;
 845	u64 start_at, mem_size;
 846
 847	if (!p)
 848		return -EINVAL;
 849
 850	if (!strncmp(p, "exactmap", 8)) {
 851#ifdef CONFIG_CRASH_DUMP
 852		/*
 853		 * If we are doing a crash dump, we still need to know
 854		 * the real mem size before original memory map is
 855		 * reset.
 856		 */
 857		saved_max_pfn = e820_end_of_ram_pfn();
 858#endif
 859		e820.nr_map = 0;
 860		userdef = 1;
 861		return 0;
 862	}
 863
 864	oldp = p;
 865	mem_size = memparse(p, &p);
 866	if (p == oldp)
 867		return -EINVAL;
 868
 869	userdef = 1;
 870	if (*p == '@') {
 871		start_at = memparse(p+1, &p);
 872		e820_add_region(start_at, mem_size, E820_RAM);
 873	} else if (*p == '#') {
 874		start_at = memparse(p+1, &p);
 875		e820_add_region(start_at, mem_size, E820_ACPI);
 876	} else if (*p == '$') {
 877		start_at = memparse(p+1, &p);
 878		e820_add_region(start_at, mem_size, E820_RESERVED);
 879	} else if (*p == '!') {
 880		start_at = memparse(p+1, &p);
 881		e820_add_region(start_at, mem_size, E820_PRAM);
 882	} else
 883		e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 884
 885	return *p == '\0' ? 0 : -EINVAL;
 886}
 
 887static int __init parse_memmap_opt(char *str)
 888{
 889	while (str) {
 890		char *k = strchr(str, ',');
 891
 892		if (k)
 893			*k++ = 0;
 894
 895		parse_memmap_one(str);
 896		str = k;
 897	}
 898
 899	return 0;
 900}
 901early_param("memmap", parse_memmap_opt);
 902
 903void __init finish_e820_parsing(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 904{
 905	if (userdef) {
 906		if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map),
 907					&e820.nr_map) < 0)
 908			early_panic("Invalid user supplied memory map");
 909
 910		printk(KERN_INFO "e820: user-defined physical RAM map:\n");
 911		e820_print_map("user");
 912	}
 913}
 914
 915static const char *e820_type_to_string(int e820_type)
 916{
 917	switch (e820_type) {
 918	case E820_RESERVED_KERN:
 919	case E820_RAM:	return "System RAM";
 920	case E820_ACPI:	return "ACPI Tables";
 921	case E820_NVS:	return "ACPI Non-volatile Storage";
 922	case E820_UNUSABLE:	return "Unusable memory";
 923	case E820_PRAM: return "Persistent Memory (legacy)";
 924	case E820_PMEM: return "Persistent Memory";
 925	default:	return "reserved";
 
 926	}
 927}
 928
 929static unsigned long e820_type_to_iomem_type(int e820_type)
 930{
 931	switch (e820_type) {
 932	case E820_RESERVED_KERN:
 933	case E820_RAM:
 934		return IORESOURCE_SYSTEM_RAM;
 935	case E820_ACPI:
 936	case E820_NVS:
 937	case E820_UNUSABLE:
 938	case E820_PRAM:
 939	case E820_PMEM:
 940	default:
 941		return IORESOURCE_MEM;
 942	}
 943}
 944
 945static unsigned long e820_type_to_iores_desc(int e820_type)
 946{
 947	switch (e820_type) {
 948	case E820_ACPI:
 949		return IORES_DESC_ACPI_TABLES;
 950	case E820_NVS:
 951		return IORES_DESC_ACPI_NV_STORAGE;
 952	case E820_PMEM:
 953		return IORES_DESC_PERSISTENT_MEMORY;
 954	case E820_PRAM:
 955		return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
 956	case E820_RESERVED_KERN:
 957	case E820_RAM:
 958	case E820_UNUSABLE:
 959	default:
 960		return IORES_DESC_NONE;
 961	}
 962}
 963
 964static bool do_mark_busy(u32 type, struct resource *res)
 965{
 966	/* this is the legacy bios/dos rom-shadow + mmio region */
 967	if (res->start < (1ULL<<20))
 968		return true;
 969
 970	/*
 971	 * Treat persistent memory like device memory, i.e. reserve it
 972	 * for exclusive use of a driver
 973	 */
 974	switch (type) {
 975	case E820_RESERVED:
 976	case E820_PRAM:
 977	case E820_PMEM:
 978		return false;
 
 
 
 
 
 979	default:
 980		return true;
 981	}
 982}
 983
 984/*
 985 * Mark e820 reserved areas as busy for the resource manager.
 986 */
 
 987static struct resource __initdata *e820_res;
 988void __init e820_reserve_resources(void)
 
 989{
 990	int i;
 991	struct resource *res;
 992	u64 end;
 993
 994	res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
 
 
 
 
 995	e820_res = res;
 996	for (i = 0; i < e820.nr_map; i++) {
 997		end = e820.map[i].addr + e820.map[i].size - 1;
 
 
 
 998		if (end != (resource_size_t)end) {
 999			res++;
1000			continue;
1001		}
1002		res->name = e820_type_to_string(e820.map[i].type);
1003		res->start = e820.map[i].addr;
1004		res->end = end;
1005
1006		res->flags = e820_type_to_iomem_type(e820.map[i].type);
1007		res->desc = e820_type_to_iores_desc(e820.map[i].type);
1008
1009		/*
1010		 * don't register the region that could be conflicted with
1011		 * pci device BAR resource and insert them later in
1012		 * pcibios_resource_survey()
1013		 */
1014		if (do_mark_busy(e820.map[i].type, res)) {
1015			res->flags |= IORESOURCE_BUSY;
1016			insert_resource(&iomem_resource, res);
1017		}
1018		res++;
1019	}
1020
1021	for (i = 0; i < e820_saved.nr_map; i++) {
1022		struct e820entry *entry = &e820_saved.map[i];
1023		firmware_map_add_early(entry->addr,
1024			entry->addr + entry->size,
1025			e820_type_to_string(entry->type));
1026	}
1027}
1028
1029/* How much should we pad RAM ending depending on where it is? */
1030static unsigned long ram_alignment(resource_size_t pos)
 
 
1031{
1032	unsigned long mb = pos >> 20;
1033
1034	/* To 64kB in the first megabyte */
1035	if (!mb)
1036		return 64*1024;
1037
1038	/* To 1MB in the first 16MB */
1039	if (mb < 16)
1040		return 1024*1024;
1041
1042	/* To 64MB for anything above that */
1043	return 64*1024*1024;
1044}
1045
1046#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1047
1048void __init e820_reserve_resources_late(void)
1049{
1050	int i;
1051	struct resource *res;
1052
1053	res = e820_res;
1054	for (i = 0; i < e820.nr_map; i++) {
1055		if (!res->parent && res->end)
1056			insert_resource_expand_to_fit(&iomem_resource, res);
1057		res++;
1058	}
1059
1060	/*
1061	 * Try to bump up RAM regions to reasonable boundaries to
1062	 * avoid stolen RAM:
1063	 */
1064	for (i = 0; i < e820.nr_map; i++) {
1065		struct e820entry *entry = &e820.map[i];
1066		u64 start, end;
1067
1068		if (entry->type != E820_RAM)
1069			continue;
 
1070		start = entry->addr + entry->size;
1071		end = round_up(start, ram_alignment(start)) - 1;
1072		if (end > MAX_RESOURCE_SIZE)
1073			end = MAX_RESOURCE_SIZE;
1074		if (start >= end)
1075			continue;
1076		printk(KERN_DEBUG
1077		       "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n",
1078		       start, end);
1079		reserve_region_with_split(&iomem_resource, start, end,
1080					  "RAM buffer");
1081	}
1082}
1083
1084char *__init default_machine_specific_memory_setup(void)
 
 
 
1085{
1086	char *who = "BIOS-e820";
1087	u32 new_nr;
1088	/*
1089	 * Try to copy the BIOS-supplied E820-map.
1090	 *
1091	 * Otherwise fake a memory map; one section from 0k->640k,
1092	 * the next section from 1mb->appropriate_mem_k
1093	 */
1094	new_nr = boot_params.e820_entries;
1095	sanitize_e820_map(boot_params.e820_map,
1096			ARRAY_SIZE(boot_params.e820_map),
1097			&new_nr);
1098	boot_params.e820_entries = new_nr;
1099	if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1100	  < 0) {
1101		u64 mem_size;
1102
1103		/* compare results from other methods and take the greater */
1104		if (boot_params.alt_mem_k
1105		    < boot_params.screen_info.ext_mem_k) {
1106			mem_size = boot_params.screen_info.ext_mem_k;
1107			who = "BIOS-88";
1108		} else {
1109			mem_size = boot_params.alt_mem_k;
1110			who = "BIOS-e801";
1111		}
1112
1113		e820.nr_map = 0;
1114		e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1115		e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1116	}
1117
1118	/* In case someone cares... */
 
 
1119	return who;
1120}
1121
1122void __init setup_memory_map(void)
 
 
 
 
 
1123{
1124	char *who;
1125
 
 
 
1126	who = x86_init.resources.memory_setup();
1127	memcpy(&e820_saved, &e820, sizeof(struct e820map));
1128	printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n");
1129	e820_print_map(who);
 
 
 
1130}
1131
1132void __init memblock_x86_fill(void)
1133{
1134	int i;
1135	u64 end;
1136
1137	/*
1138	 * EFI may have more than 128 entries
1139	 * We are safe to enable resizing, beause memblock_x86_fill()
1140	 * is rather later for x86
 
 
 
 
1141	 */
1142	memblock_allow_resize();
1143
1144	for (i = 0; i < e820.nr_map; i++) {
1145		struct e820entry *ei = &e820.map[i];
1146
1147		end = ei->addr + ei->size;
1148		if (end != (resource_size_t)end)
1149			continue;
1150
1151		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1152			continue;
1153
1154		memblock_add(ei->addr, ei->size);
1155	}
1156
1157	/* throw away partial pages */
1158	memblock_trim_memory(PAGE_SIZE);
1159
1160	memblock_dump_all();
1161}
1162
1163void __init memblock_find_dma_reserve(void)
1164{
1165#ifdef CONFIG_X86_64
1166	u64 nr_pages = 0, nr_free_pages = 0;
1167	unsigned long start_pfn, end_pfn;
1168	phys_addr_t start, end;
1169	int i;
1170	u64 u;
1171
1172	/*
1173	 * need to find out used area below MAX_DMA_PFN
1174	 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1175	 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1176	 */
1177	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1178		start_pfn = min(start_pfn, MAX_DMA_PFN);
1179		end_pfn = min(end_pfn, MAX_DMA_PFN);
1180		nr_pages += end_pfn - start_pfn;
1181	}
1182
1183	for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1184				NULL) {
1185		start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1186		end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1187		if (start_pfn < end_pfn)
1188			nr_free_pages += end_pfn - start_pfn;
1189	}
1190
1191	set_dma_reserve(nr_pages - nr_free_pages);
1192#endif
1193}