Linux Audio

Check our new training course

Loading...
v3.1
   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/bootmem.h>
  16#include <linux/pfn.h>
  17#include <linux/suspend.h>
  18#include <linux/acpi.h>
  19#include <linux/firmware-map.h>
  20#include <linux/memblock.h>
 
  21
  22#include <asm/e820.h>
  23#include <asm/proto.h>
  24#include <asm/setup.h>
 
  25
  26/*
  27 * The e820 map is the map that gets modified e.g. with command line parameters
  28 * and that is also registered with modifications in the kernel resource tree
  29 * with the iomem_resource as parent.
  30 *
  31 * The e820_saved is directly saved after the BIOS-provided memory map is
  32 * copied. It doesn't get modified afterwards. It's registered for the
  33 * /sys/firmware/memmap interface.
  34 *
  35 * That memory map is not modified and is used as base for kexec. The kexec'd
  36 * kernel should get the same memory map as the firmware provides. Then the
  37 * user can e.g. boot the original kernel with mem=1G while still booting the
  38 * next kernel with full memory.
  39 */
  40struct e820map e820;
  41struct e820map e820_saved;
 
 
  42
  43/* For PCI or other memory-mapped resources */
  44unsigned long pci_mem_start = 0xaeedbabe;
  45#ifdef CONFIG_PCI
  46EXPORT_SYMBOL(pci_mem_start);
  47#endif
  48
  49/*
  50 * This function checks if any part of the range <start,end> is mapped
  51 * with type.
  52 */
  53int
  54e820_any_mapped(u64 start, u64 end, unsigned type)
  55{
  56	int i;
  57
  58	for (i = 0; i < e820.nr_map; i++) {
  59		struct e820entry *ei = &e820.map[i];
  60
  61		if (type && ei->type != type)
  62			continue;
  63		if (ei->addr >= end || ei->addr + ei->size <= start)
  64			continue;
  65		return 1;
  66	}
  67	return 0;
  68}
  69EXPORT_SYMBOL_GPL(e820_any_mapped);
  70
  71/*
  72 * This function checks if the entire range <start,end> is mapped with type.
  73 *
  74 * Note: this function only works correct if the e820 table is sorted and
  75 * not-overlapping, which is the case
  76 */
  77int __init e820_all_mapped(u64 start, u64 end, unsigned type)
  78{
  79	int i;
  80
  81	for (i = 0; i < e820.nr_map; i++) {
  82		struct e820entry *ei = &e820.map[i];
  83
  84		if (type && ei->type != type)
  85			continue;
  86		/* is the region (part) in overlap with the current region ?*/
  87		if (ei->addr >= end || ei->addr + ei->size <= start)
  88			continue;
  89
  90		/* if the region is at the beginning of <start,end> we move
  91		 * start to the end of the region since it's ok until there
  92		 */
  93		if (ei->addr <= start)
  94			start = ei->addr + ei->size;
  95		/*
  96		 * if start is now at or beyond end, we're done, full
  97		 * coverage
  98		 */
  99		if (start >= end)
 100			return 1;
 101	}
 102	return 0;
 103}
 104
 105/*
 106 * Add a memory region to the kernel e820 map.
 107 */
 108static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
 109					 int type)
 110{
 111	int x = e820x->nr_map;
 112
 113	if (x >= ARRAY_SIZE(e820x->map)) {
 114		printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
 
 
 115		return;
 116	}
 117
 118	e820x->map[x].addr = start;
 119	e820x->map[x].size = size;
 120	e820x->map[x].type = type;
 121	e820x->nr_map++;
 122}
 123
 124void __init e820_add_region(u64 start, u64 size, int type)
 125{
 126	__e820_add_region(&e820, start, size, type);
 127}
 128
 129static void __init e820_print_type(u32 type)
 130{
 131	switch (type) {
 132	case E820_RAM:
 133	case E820_RESERVED_KERN:
 134		printk(KERN_CONT "(usable)");
 135		break;
 136	case E820_RESERVED:
 137		printk(KERN_CONT "(reserved)");
 138		break;
 139	case E820_ACPI:
 140		printk(KERN_CONT "(ACPI data)");
 141		break;
 142	case E820_NVS:
 143		printk(KERN_CONT "(ACPI NVS)");
 144		break;
 145	case E820_UNUSABLE:
 146		printk(KERN_CONT "(unusable)");
 
 
 
 
 147		break;
 148	default:
 149		printk(KERN_CONT "type %u", type);
 150		break;
 151	}
 152}
 153
 154void __init e820_print_map(char *who)
 155{
 156	int i;
 157
 158	for (i = 0; i < e820.nr_map; i++) {
 159		printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
 160		       (unsigned long long) e820.map[i].addr,
 161		       (unsigned long long)
 162		       (e820.map[i].addr + e820.map[i].size));
 163		e820_print_type(e820.map[i].type);
 164		printk(KERN_CONT "\n");
 165	}
 166}
 167
 168/*
 169 * Sanitize the BIOS e820 map.
 170 *
 171 * Some e820 responses include overlapping entries. The following
 172 * replaces the original e820 map with a new one, removing overlaps,
 173 * and resolving conflicting memory types in favor of highest
 174 * numbered type.
 175 *
 176 * The input parameter biosmap points to an array of 'struct
 177 * e820entry' which on entry has elements in the range [0, *pnr_map)
 178 * valid, and which has space for up to max_nr_map entries.
 179 * On return, the resulting sanitized e820 map entries will be in
 180 * overwritten in the same location, starting at biosmap.
 181 *
 182 * The integer pointed to by pnr_map must be valid on entry (the
 183 * current number of valid entries located at biosmap) and will
 184 * be updated on return, with the new number of valid entries
 185 * (something no more than max_nr_map.)
 186 *
 187 * The return value from sanitize_e820_map() is zero if it
 188 * successfully 'sanitized' the map entries passed in, and is -1
 189 * if it did nothing, which can happen if either of (1) it was
 190 * only passed one map entry, or (2) any of the input map entries
 191 * were invalid (start + size < start, meaning that the size was
 192 * so big the described memory range wrapped around through zero.)
 193 *
 194 *	Visually we're performing the following
 195 *	(1,2,3,4 = memory types)...
 196 *
 197 *	Sample memory map (w/overlaps):
 198 *	   ____22__________________
 199 *	   ______________________4_
 200 *	   ____1111________________
 201 *	   _44_____________________
 202 *	   11111111________________
 203 *	   ____________________33__
 204 *	   ___________44___________
 205 *	   __________33333_________
 206 *	   ______________22________
 207 *	   ___________________2222_
 208 *	   _________111111111______
 209 *	   _____________________11_
 210 *	   _________________4______
 211 *
 212 *	Sanitized equivalent (no overlap):
 213 *	   1_______________________
 214 *	   _44_____________________
 215 *	   ___1____________________
 216 *	   ____22__________________
 217 *	   ______11________________
 218 *	   _________1______________
 219 *	   __________3_____________
 220 *	   ___________44___________
 221 *	   _____________33_________
 222 *	   _______________2________
 223 *	   ________________1_______
 224 *	   _________________4______
 225 *	   ___________________2____
 226 *	   ____________________33__
 227 *	   ______________________4_
 228 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 229
 230int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
 231			     u32 *pnr_map)
 232{
 233	struct change_member {
 234		struct e820entry *pbios; /* pointer to original bios entry */
 235		unsigned long long addr; /* address for this change point */
 236	};
 237	static struct change_member change_point_list[2*E820_X_MAX] __initdata;
 238	static struct change_member *change_point[2*E820_X_MAX] __initdata;
 239	static struct e820entry *overlap_list[E820_X_MAX] __initdata;
 240	static struct e820entry new_bios[E820_X_MAX] __initdata;
 241	struct change_member *change_tmp;
 242	unsigned long current_type, last_type;
 243	unsigned long long last_addr;
 244	int chgidx, still_changing;
 245	int overlap_entries;
 246	int new_bios_entry;
 247	int old_nr, new_nr, chg_nr;
 248	int i;
 249
 250	/* if there's only one memory region, don't bother */
 251	if (*pnr_map < 2)
 252		return -1;
 253
 254	old_nr = *pnr_map;
 255	BUG_ON(old_nr > max_nr_map);
 256
 257	/* bail out if we find any unreasonable addresses in bios map */
 258	for (i = 0; i < old_nr; i++)
 259		if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 260			return -1;
 261
 262	/* create pointers for initial change-point information (for sorting) */
 263	for (i = 0; i < 2 * old_nr; i++)
 264		change_point[i] = &change_point_list[i];
 265
 266	/* record all known change-points (starting and ending addresses),
 267	   omitting those that are for empty memory regions */
 268	chgidx = 0;
 269	for (i = 0; i < old_nr; i++)	{
 270		if (biosmap[i].size != 0) {
 271			change_point[chgidx]->addr = biosmap[i].addr;
 272			change_point[chgidx++]->pbios = &biosmap[i];
 273			change_point[chgidx]->addr = biosmap[i].addr +
 274				biosmap[i].size;
 275			change_point[chgidx++]->pbios = &biosmap[i];
 276		}
 277	}
 278	chg_nr = chgidx;
 279
 280	/* sort change-point list by memory addresses (low -> high) */
 281	still_changing = 1;
 282	while (still_changing)	{
 283		still_changing = 0;
 284		for (i = 1; i < chg_nr; i++)  {
 285			unsigned long long curaddr, lastaddr;
 286			unsigned long long curpbaddr, lastpbaddr;
 287
 288			curaddr = change_point[i]->addr;
 289			lastaddr = change_point[i - 1]->addr;
 290			curpbaddr = change_point[i]->pbios->addr;
 291			lastpbaddr = change_point[i - 1]->pbios->addr;
 292
 293			/*
 294			 * swap entries, when:
 295			 *
 296			 * curaddr > lastaddr or
 297			 * curaddr == lastaddr and curaddr == curpbaddr and
 298			 * lastaddr != lastpbaddr
 299			 */
 300			if (curaddr < lastaddr ||
 301			    (curaddr == lastaddr && curaddr == curpbaddr &&
 302			     lastaddr != lastpbaddr)) {
 303				change_tmp = change_point[i];
 304				change_point[i] = change_point[i-1];
 305				change_point[i-1] = change_tmp;
 306				still_changing = 1;
 307			}
 308		}
 309	}
 310
 311	/* create a new bios memory map, removing overlaps */
 312	overlap_entries = 0;	 /* number of entries in the overlap table */
 313	new_bios_entry = 0;	 /* index for creating new bios map entries */
 314	last_type = 0;		 /* start with undefined memory type */
 315	last_addr = 0;		 /* start with 0 as last starting address */
 316
 317	/* loop through change-points, determining affect on the new bios map */
 318	for (chgidx = 0; chgidx < chg_nr; chgidx++) {
 319		/* keep track of all overlapping bios entries */
 320		if (change_point[chgidx]->addr ==
 321		    change_point[chgidx]->pbios->addr) {
 322			/*
 323			 * add map entry to overlap list (> 1 entry
 324			 * implies an overlap)
 325			 */
 326			overlap_list[overlap_entries++] =
 327				change_point[chgidx]->pbios;
 328		} else {
 329			/*
 330			 * remove entry from list (order independent,
 331			 * so swap with last)
 332			 */
 333			for (i = 0; i < overlap_entries; i++) {
 334				if (overlap_list[i] ==
 335				    change_point[chgidx]->pbios)
 336					overlap_list[i] =
 337						overlap_list[overlap_entries-1];
 338			}
 339			overlap_entries--;
 340		}
 341		/*
 342		 * if there are overlapping entries, decide which
 343		 * "type" to use (larger value takes precedence --
 344		 * 1=usable, 2,3,4,4+=unusable)
 345		 */
 346		current_type = 0;
 347		for (i = 0; i < overlap_entries; i++)
 348			if (overlap_list[i]->type > current_type)
 349				current_type = overlap_list[i]->type;
 350		/*
 351		 * continue building up new bios map based on this
 352		 * information
 353		 */
 354		if (current_type != last_type)	{
 355			if (last_type != 0)	 {
 356				new_bios[new_bios_entry].size =
 357					change_point[chgidx]->addr - last_addr;
 358				/*
 359				 * move forward only if the new size
 360				 * was non-zero
 361				 */
 362				if (new_bios[new_bios_entry].size != 0)
 363					/*
 364					 * no more space left for new
 365					 * bios entries ?
 366					 */
 367					if (++new_bios_entry >= max_nr_map)
 368						break;
 369			}
 370			if (current_type != 0)	{
 371				new_bios[new_bios_entry].addr =
 372					change_point[chgidx]->addr;
 373				new_bios[new_bios_entry].type = current_type;
 374				last_addr = change_point[chgidx]->addr;
 375			}
 376			last_type = current_type;
 377		}
 378	}
 379	/* retain count for new bios entries */
 380	new_nr = new_bios_entry;
 381
 382	/* copy new bios mapping into original location */
 383	memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
 384	*pnr_map = new_nr;
 385
 386	return 0;
 387}
 388
 389static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
 390{
 391	while (nr_map) {
 392		u64 start = biosmap->addr;
 393		u64 size = biosmap->size;
 394		u64 end = start + size;
 395		u32 type = biosmap->type;
 396
 397		/* Overflow in 64 bits? Ignore the memory map. */
 398		if (start > end)
 399			return -1;
 400
 401		e820_add_region(start, size, type);
 402
 403		biosmap++;
 404		nr_map--;
 405	}
 406	return 0;
 407}
 408
 409/*
 410 * Copy the BIOS e820 map into a safe place.
 411 *
 412 * Sanity-check it while we're at it..
 413 *
 414 * If we're lucky and live on a modern system, the setup code
 415 * will have given us a memory map that we can use to properly
 416 * set up memory.  If we aren't, we'll fake a memory map.
 417 */
 418static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
 419{
 420	/* Only one memory region (or negative)? Ignore it */
 421	if (nr_map < 2)
 422		return -1;
 423
 424	return __append_e820_map(biosmap, nr_map);
 425}
 426
 427static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
 428					u64 size, unsigned old_type,
 429					unsigned new_type)
 430{
 431	u64 end;
 432	unsigned int i;
 433	u64 real_updated_size = 0;
 434
 435	BUG_ON(old_type == new_type);
 436
 437	if (size > (ULLONG_MAX - start))
 438		size = ULLONG_MAX - start;
 439
 440	end = start + size;
 441	printk(KERN_DEBUG "e820 update range: %016Lx - %016Lx ",
 442		       (unsigned long long) start,
 443		       (unsigned long long) end);
 444	e820_print_type(old_type);
 445	printk(KERN_CONT " ==> ");
 446	e820_print_type(new_type);
 447	printk(KERN_CONT "\n");
 448
 449	for (i = 0; i < e820x->nr_map; i++) {
 450		struct e820entry *ei = &e820x->map[i];
 451		u64 final_start, final_end;
 452		u64 ei_end;
 453
 454		if (ei->type != old_type)
 455			continue;
 456
 457		ei_end = ei->addr + ei->size;
 458		/* totally covered by new range? */
 459		if (ei->addr >= start && ei_end <= end) {
 460			ei->type = new_type;
 461			real_updated_size += ei->size;
 462			continue;
 463		}
 464
 465		/* new range is totally covered? */
 466		if (ei->addr < start && ei_end > end) {
 467			__e820_add_region(e820x, start, size, new_type);
 468			__e820_add_region(e820x, end, ei_end - end, ei->type);
 469			ei->size = start - ei->addr;
 470			real_updated_size += size;
 471			continue;
 472		}
 473
 474		/* partially covered */
 475		final_start = max(start, ei->addr);
 476		final_end = min(end, ei_end);
 477		if (final_start >= final_end)
 478			continue;
 479
 480		__e820_add_region(e820x, final_start, final_end - final_start,
 481				  new_type);
 482
 483		real_updated_size += final_end - final_start;
 484
 485		/*
 486		 * left range could be head or tail, so need to update
 487		 * size at first.
 488		 */
 489		ei->size -= final_end - final_start;
 490		if (ei->addr < final_start)
 491			continue;
 492		ei->addr = final_end;
 493	}
 494	return real_updated_size;
 495}
 496
 497u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
 498			     unsigned new_type)
 499{
 500	return __e820_update_range(&e820, start, size, old_type, new_type);
 501}
 502
 503static u64 __init e820_update_range_saved(u64 start, u64 size,
 504					  unsigned old_type, unsigned new_type)
 505{
 506	return __e820_update_range(&e820_saved, start, size, old_type,
 507				     new_type);
 508}
 509
 510/* make e820 not cover the range */
 511u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
 512			     int checktype)
 513{
 514	int i;
 515	u64 end;
 516	u64 real_removed_size = 0;
 517
 518	if (size > (ULLONG_MAX - start))
 519		size = ULLONG_MAX - start;
 520
 521	end = start + size;
 522	printk(KERN_DEBUG "e820 remove range: %016Lx - %016Lx ",
 523		       (unsigned long long) start,
 524		       (unsigned long long) end);
 525	if (checktype)
 526		e820_print_type(old_type);
 527	printk(KERN_CONT "\n");
 528
 529	for (i = 0; i < e820.nr_map; i++) {
 530		struct e820entry *ei = &e820.map[i];
 531		u64 final_start, final_end;
 532		u64 ei_end;
 533
 534		if (checktype && ei->type != old_type)
 535			continue;
 536
 537		ei_end = ei->addr + ei->size;
 538		/* totally covered? */
 539		if (ei->addr >= start && ei_end <= end) {
 540			real_removed_size += ei->size;
 541			memset(ei, 0, sizeof(struct e820entry));
 542			continue;
 543		}
 544
 545		/* new range is totally covered? */
 546		if (ei->addr < start && ei_end > end) {
 547			e820_add_region(end, ei_end - end, ei->type);
 548			ei->size = start - ei->addr;
 549			real_removed_size += size;
 550			continue;
 551		}
 552
 553		/* partially covered */
 554		final_start = max(start, ei->addr);
 555		final_end = min(end, ei_end);
 556		if (final_start >= final_end)
 557			continue;
 558		real_removed_size += final_end - final_start;
 559
 560		/*
 561		 * left range could be head or tail, so need to update
 562		 * size at first.
 563		 */
 564		ei->size -= final_end - final_start;
 565		if (ei->addr < final_start)
 566			continue;
 567		ei->addr = final_end;
 568	}
 569	return real_removed_size;
 570}
 571
 572void __init update_e820(void)
 573{
 574	u32 nr_map;
 575
 576	nr_map = e820.nr_map;
 577	if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
 578		return;
 579	e820.nr_map = nr_map;
 580	printk(KERN_INFO "modified physical RAM map:\n");
 581	e820_print_map("modified");
 582}
 583static void __init update_e820_saved(void)
 584{
 585	u32 nr_map;
 586
 587	nr_map = e820_saved.nr_map;
 588	if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
 589		return;
 590	e820_saved.nr_map = nr_map;
 591}
 592#define MAX_GAP_END 0x100000000ull
 593/*
 594 * Search for a gap in the e820 memory space from start_addr to end_addr.
 595 */
 596__init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
 597		unsigned long start_addr, unsigned long long end_addr)
 598{
 599	unsigned long long last;
 600	int i = e820.nr_map;
 601	int found = 0;
 602
 603	last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
 604
 605	while (--i >= 0) {
 606		unsigned long long start = e820.map[i].addr;
 607		unsigned long long end = start + e820.map[i].size;
 608
 609		if (end < start_addr)
 610			continue;
 611
 612		/*
 613		 * Since "last" is at most 4GB, we know we'll
 614		 * fit in 32 bits if this condition is true
 615		 */
 616		if (last > end) {
 617			unsigned long gap = last - end;
 618
 619			if (gap >= *gapsize) {
 620				*gapsize = gap;
 621				*gapstart = end;
 622				found = 1;
 623			}
 624		}
 625		if (start < last)
 626			last = start;
 627	}
 628	return found;
 629}
 630
 631/*
 632 * Search for the biggest gap in the low 32 bits of the e820
 633 * memory space.  We pass this space to PCI to assign MMIO resources
 634 * for hotplug or unconfigured devices in.
 635 * Hopefully the BIOS let enough space left.
 636 */
 637__init void e820_setup_gap(void)
 638{
 639	unsigned long gapstart, gapsize;
 640	int found;
 641
 642	gapstart = 0x10000000;
 643	gapsize = 0x400000;
 644	found  = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
 645
 646#ifdef CONFIG_X86_64
 647	if (!found) {
 648		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
 649		printk(KERN_ERR
 650	"PCI: Warning: Cannot find a gap in the 32bit address range\n"
 651	"PCI: Unassigned devices with 32bit resource registers may break!\n");
 652	}
 653#endif
 654
 655	/*
 656	 * e820_reserve_resources_late protect stolen RAM already
 657	 */
 658	pci_mem_start = gapstart;
 659
 660	printk(KERN_INFO
 661	       "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
 662	       pci_mem_start, gapstart, gapsize);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 663}
 664
 665/**
 666 * Because of the size limitation of struct boot_params, only first
 667 * 128 E820 memory entries are passed to kernel via
 668 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
 669 * linked list of struct setup_data, which is parsed here.
 670 */
 671void __init parse_e820_ext(struct setup_data *sdata)
 672{
 673	int entries;
 674	struct e820entry *extmap;
 
 675
 
 676	entries = sdata->len / sizeof(struct e820entry);
 677	extmap = (struct e820entry *)(sdata->data);
 678	__append_e820_map(extmap, entries);
 679	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 680	printk(KERN_INFO "extended physical RAM map:\n");
 
 681	e820_print_map("extended");
 682}
 683
 684#if defined(CONFIG_X86_64) || \
 685	(defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
 686/**
 687 * Find the ranges of physical addresses that do not correspond to
 688 * e820 RAM areas and mark the corresponding pages as nosave for
 689 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
 690 *
 691 * This function requires the e820 map to be sorted and without any
 692 * overlapping entries and assumes the first e820 area to be RAM.
 693 */
 694void __init e820_mark_nosave_regions(unsigned long limit_pfn)
 695{
 696	int i;
 697	unsigned long pfn;
 698
 699	pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
 700	for (i = 1; i < e820.nr_map; i++) {
 701		struct e820entry *ei = &e820.map[i];
 702
 703		if (pfn < PFN_UP(ei->addr))
 704			register_nosave_region(pfn, PFN_UP(ei->addr));
 705
 706		pfn = PFN_DOWN(ei->addr + ei->size);
 
 707		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
 708			register_nosave_region(PFN_UP(ei->addr), pfn);
 709
 710		if (pfn >= limit_pfn)
 711			break;
 712	}
 713}
 714#endif
 715
 716#ifdef CONFIG_HIBERNATION
 717/**
 718 * Mark ACPI NVS memory region, so that we can save/restore it during
 719 * hibernation and the subsequent resume.
 720 */
 721static int __init e820_mark_nvs_memory(void)
 722{
 723	int i;
 724
 725	for (i = 0; i < e820.nr_map; i++) {
 726		struct e820entry *ei = &e820.map[i];
 727
 728		if (ei->type == E820_NVS)
 729			suspend_nvs_register(ei->addr, ei->size);
 730	}
 731
 732	return 0;
 733}
 734core_initcall(e820_mark_nvs_memory);
 735#endif
 736
 737/*
 738 * pre allocated 4k and reserved it in memblock and e820_saved
 739 */
 740u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
 741{
 742	u64 size = 0;
 743	u64 addr;
 744	u64 start;
 745
 746	for (start = startt; ; start += size) {
 747		start = memblock_x86_find_in_range_size(start, &size, align);
 748		if (start == MEMBLOCK_ERROR)
 749			return 0;
 750		if (size >= sizet)
 751			break;
 752	}
 753
 754#ifdef CONFIG_X86_32
 755	if (start >= MAXMEM)
 756		return 0;
 757	if (start + size > MAXMEM)
 758		size = MAXMEM - start;
 759#endif
 760
 761	addr = round_down(start + size - sizet, align);
 762	if (addr < start)
 763		return 0;
 764	memblock_x86_reserve_range(addr, addr + sizet, "new next");
 765	e820_update_range_saved(addr, sizet, E820_RAM, E820_RESERVED);
 766	printk(KERN_INFO "update e820_saved for early_reserve_e820\n");
 767	update_e820_saved();
 768
 769	return addr;
 770}
 771
 772#ifdef CONFIG_X86_32
 773# ifdef CONFIG_X86_PAE
 774#  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
 775# else
 776#  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
 777# endif
 778#else /* CONFIG_X86_32 */
 779# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
 780#endif
 781
 782/*
 783 * Find the highest page frame number we have available
 784 */
 785static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
 786{
 787	int i;
 788	unsigned long last_pfn = 0;
 789	unsigned long max_arch_pfn = MAX_ARCH_PFN;
 790
 791	for (i = 0; i < e820.nr_map; i++) {
 792		struct e820entry *ei = &e820.map[i];
 793		unsigned long start_pfn;
 794		unsigned long end_pfn;
 795
 796		if (ei->type != type)
 797			continue;
 798
 799		start_pfn = ei->addr >> PAGE_SHIFT;
 800		end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
 801
 802		if (start_pfn >= limit_pfn)
 803			continue;
 804		if (end_pfn > limit_pfn) {
 805			last_pfn = limit_pfn;
 806			break;
 807		}
 808		if (end_pfn > last_pfn)
 809			last_pfn = end_pfn;
 810	}
 811
 812	if (last_pfn > max_arch_pfn)
 813		last_pfn = max_arch_pfn;
 814
 815	printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
 816			 last_pfn, max_arch_pfn);
 817	return last_pfn;
 818}
 819unsigned long __init e820_end_of_ram_pfn(void)
 820{
 821	return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
 822}
 823
 824unsigned long __init e820_end_of_low_ram_pfn(void)
 825{
 826	return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
 827}
 828
 829static void early_panic(char *msg)
 830{
 831	early_printk(msg);
 832	panic(msg);
 833}
 834
 835static int userdef __initdata;
 836
 837/* "mem=nopentium" disables the 4MB page tables. */
 838static int __init parse_memopt(char *p)
 839{
 840	u64 mem_size;
 841
 842	if (!p)
 843		return -EINVAL;
 844
 845	if (!strcmp(p, "nopentium")) {
 846#ifdef CONFIG_X86_32
 847		setup_clear_cpu_cap(X86_FEATURE_PSE);
 848		return 0;
 849#else
 850		printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
 851		return -EINVAL;
 852#endif
 853	}
 854
 855	userdef = 1;
 856	mem_size = memparse(p, &p);
 857	/* don't remove all of memory when handling "mem={invalid}" param */
 858	if (mem_size == 0)
 859		return -EINVAL;
 860	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
 861
 862	return 0;
 863}
 864early_param("mem", parse_memopt);
 865
 866static int __init parse_memmap_opt(char *p)
 867{
 868	char *oldp;
 869	u64 start_at, mem_size;
 870
 871	if (!p)
 872		return -EINVAL;
 873
 874	if (!strncmp(p, "exactmap", 8)) {
 875#ifdef CONFIG_CRASH_DUMP
 876		/*
 877		 * If we are doing a crash dump, we still need to know
 878		 * the real mem size before original memory map is
 879		 * reset.
 880		 */
 881		saved_max_pfn = e820_end_of_ram_pfn();
 882#endif
 883		e820.nr_map = 0;
 884		userdef = 1;
 885		return 0;
 886	}
 887
 888	oldp = p;
 889	mem_size = memparse(p, &p);
 890	if (p == oldp)
 891		return -EINVAL;
 892
 893	userdef = 1;
 894	if (*p == '@') {
 895		start_at = memparse(p+1, &p);
 896		e820_add_region(start_at, mem_size, E820_RAM);
 897	} else if (*p == '#') {
 898		start_at = memparse(p+1, &p);
 899		e820_add_region(start_at, mem_size, E820_ACPI);
 900	} else if (*p == '$') {
 901		start_at = memparse(p+1, &p);
 902		e820_add_region(start_at, mem_size, E820_RESERVED);
 
 
 
 903	} else
 904		e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
 905
 906	return *p == '\0' ? 0 : -EINVAL;
 907}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 908early_param("memmap", parse_memmap_opt);
 909
 910void __init finish_e820_parsing(void)
 911{
 912	if (userdef) {
 913		u32 nr = e820.nr_map;
 914
 915		if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
 916			early_panic("Invalid user supplied memory map");
 917		e820.nr_map = nr;
 918
 919		printk(KERN_INFO "user-defined physical RAM map:\n");
 920		e820_print_map("user");
 921	}
 922}
 923
 924static inline const char *e820_type_to_string(int e820_type)
 925{
 926	switch (e820_type) {
 927	case E820_RESERVED_KERN:
 928	case E820_RAM:	return "System RAM";
 929	case E820_ACPI:	return "ACPI Tables";
 930	case E820_NVS:	return "ACPI Non-volatile Storage";
 931	case E820_UNUSABLE:	return "Unusable memory";
 
 
 932	default:	return "reserved";
 933	}
 934}
 935
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 936/*
 937 * Mark e820 reserved areas as busy for the resource manager.
 938 */
 939static struct resource __initdata *e820_res;
 940void __init e820_reserve_resources(void)
 941{
 942	int i;
 943	struct resource *res;
 944	u64 end;
 945
 946	res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
 947	e820_res = res;
 948	for (i = 0; i < e820.nr_map; i++) {
 949		end = e820.map[i].addr + e820.map[i].size - 1;
 950		if (end != (resource_size_t)end) {
 951			res++;
 952			continue;
 953		}
 954		res->name = e820_type_to_string(e820.map[i].type);
 955		res->start = e820.map[i].addr;
 956		res->end = end;
 957
 958		res->flags = IORESOURCE_MEM;
 
 959
 960		/*
 961		 * don't register the region that could be conflicted with
 962		 * pci device BAR resource and insert them later in
 963		 * pcibios_resource_survey()
 964		 */
 965		if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20)) {
 966			res->flags |= IORESOURCE_BUSY;
 967			insert_resource(&iomem_resource, res);
 968		}
 969		res++;
 970	}
 971
 972	for (i = 0; i < e820_saved.nr_map; i++) {
 973		struct e820entry *entry = &e820_saved.map[i];
 974		firmware_map_add_early(entry->addr,
 975			entry->addr + entry->size - 1,
 976			e820_type_to_string(entry->type));
 977	}
 978}
 979
 980/* How much should we pad RAM ending depending on where it is? */
 981static unsigned long ram_alignment(resource_size_t pos)
 982{
 983	unsigned long mb = pos >> 20;
 984
 985	/* To 64kB in the first megabyte */
 986	if (!mb)
 987		return 64*1024;
 988
 989	/* To 1MB in the first 16MB */
 990	if (mb < 16)
 991		return 1024*1024;
 992
 993	/* To 64MB for anything above that */
 994	return 64*1024*1024;
 995}
 996
 997#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
 998
 999void __init e820_reserve_resources_late(void)
1000{
1001	int i;
1002	struct resource *res;
1003
1004	res = e820_res;
1005	for (i = 0; i < e820.nr_map; i++) {
1006		if (!res->parent && res->end)
1007			insert_resource_expand_to_fit(&iomem_resource, res);
1008		res++;
1009	}
1010
1011	/*
1012	 * Try to bump up RAM regions to reasonable boundaries to
1013	 * avoid stolen RAM:
1014	 */
1015	for (i = 0; i < e820.nr_map; i++) {
1016		struct e820entry *entry = &e820.map[i];
1017		u64 start, end;
1018
1019		if (entry->type != E820_RAM)
1020			continue;
1021		start = entry->addr + entry->size;
1022		end = round_up(start, ram_alignment(start)) - 1;
1023		if (end > MAX_RESOURCE_SIZE)
1024			end = MAX_RESOURCE_SIZE;
1025		if (start >= end)
1026			continue;
1027		printk(KERN_DEBUG "reserve RAM buffer: %016llx - %016llx ",
1028			       start, end);
 
1029		reserve_region_with_split(&iomem_resource, start, end,
1030					  "RAM buffer");
1031	}
1032}
1033
1034char *__init default_machine_specific_memory_setup(void)
1035{
1036	char *who = "BIOS-e820";
1037	u32 new_nr;
1038	/*
1039	 * Try to copy the BIOS-supplied E820-map.
1040	 *
1041	 * Otherwise fake a memory map; one section from 0k->640k,
1042	 * the next section from 1mb->appropriate_mem_k
1043	 */
1044	new_nr = boot_params.e820_entries;
1045	sanitize_e820_map(boot_params.e820_map,
1046			ARRAY_SIZE(boot_params.e820_map),
1047			&new_nr);
1048	boot_params.e820_entries = new_nr;
1049	if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1050	  < 0) {
1051		u64 mem_size;
1052
1053		/* compare results from other methods and take the greater */
1054		if (boot_params.alt_mem_k
1055		    < boot_params.screen_info.ext_mem_k) {
1056			mem_size = boot_params.screen_info.ext_mem_k;
1057			who = "BIOS-88";
1058		} else {
1059			mem_size = boot_params.alt_mem_k;
1060			who = "BIOS-e801";
1061		}
1062
1063		e820.nr_map = 0;
1064		e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1065		e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1066	}
1067
1068	/* In case someone cares... */
1069	return who;
1070}
1071
1072void __init setup_memory_map(void)
1073{
1074	char *who;
1075
1076	who = x86_init.resources.memory_setup();
1077	memcpy(&e820_saved, &e820, sizeof(struct e820map));
1078	printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1079	e820_print_map(who);
1080}
1081
1082void __init memblock_x86_fill(void)
1083{
1084	int i;
1085	u64 end;
1086
1087	/*
1088	 * EFI may have more than 128 entries
1089	 * We are safe to enable resizing, beause memblock_x86_fill()
1090	 * is rather later for x86
1091	 */
1092	memblock_can_resize = 1;
1093
1094	for (i = 0; i < e820.nr_map; i++) {
1095		struct e820entry *ei = &e820.map[i];
1096
1097		end = ei->addr + ei->size;
1098		if (end != (resource_size_t)end)
1099			continue;
1100
1101		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1102			continue;
1103
1104		memblock_add(ei->addr, ei->size);
1105	}
1106
1107	memblock_analyze();
 
 
1108	memblock_dump_all();
1109}
1110
1111void __init memblock_find_dma_reserve(void)
1112{
1113#ifdef CONFIG_X86_64
1114	u64 free_size_pfn;
1115	u64 mem_size_pfn;
 
 
 
 
1116	/*
1117	 * need to find out used area below MAX_DMA_PFN
1118	 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1119	 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1120	 */
1121	mem_size_pfn = memblock_x86_memory_in_range(0, MAX_DMA_PFN << PAGE_SHIFT) >> PAGE_SHIFT;
1122	free_size_pfn = memblock_x86_free_memory_in_range(0, MAX_DMA_PFN << PAGE_SHIFT) >> PAGE_SHIFT;
1123	set_dma_reserve(mem_size_pfn - free_size_pfn);
 
 
 
 
 
 
 
 
 
 
 
 
1124#endif
1125}
v4.10.11
   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 */
  43static struct e820map initial_e820  __initdata;
  44static struct e820map initial_e820_saved  __initdata;
  45struct e820map *e820 __refdata = &initial_e820;
  46struct e820map *e820_saved __refdata = &initial_e820_saved;
  47
  48/* For PCI or other memory-mapped resources */
  49unsigned long pci_mem_start = 0xaeedbabe;
  50#ifdef CONFIG_PCI
  51EXPORT_SYMBOL(pci_mem_start);
  52#endif
  53
  54/*
  55 * This function checks if any part of the range <start,end> is mapped
  56 * with type.
  57 */
  58int
  59e820_any_mapped(u64 start, u64 end, unsigned type)
  60{
  61	int i;
  62
  63	for (i = 0; i < e820->nr_map; i++) {
  64		struct e820entry *ei = &e820->map[i];
  65
  66		if (type && ei->type != type)
  67			continue;
  68		if (ei->addr >= end || ei->addr + ei->size <= start)
  69			continue;
  70		return 1;
  71	}
  72	return 0;
  73}
  74EXPORT_SYMBOL_GPL(e820_any_mapped);
  75
  76/*
  77 * This function checks if the entire range <start,end> is mapped with type.
  78 *
  79 * Note: this function only works correct if the e820 table is sorted and
  80 * not-overlapping, which is the case
  81 */
  82int __init e820_all_mapped(u64 start, u64 end, unsigned type)
  83{
  84	int i;
  85
  86	for (i = 0; i < e820->nr_map; i++) {
  87		struct e820entry *ei = &e820->map[i];
  88
  89		if (type && ei->type != type)
  90			continue;
  91		/* is the region (part) in overlap with the current region ?*/
  92		if (ei->addr >= end || ei->addr + ei->size <= start)
  93			continue;
  94
  95		/* if the region is at the beginning of <start,end> we move
  96		 * start to the end of the region since it's ok until there
  97		 */
  98		if (ei->addr <= start)
  99			start = ei->addr + ei->size;
 100		/*
 101		 * if start is now at or beyond end, we're done, full
 102		 * coverage
 103		 */
 104		if (start >= end)
 105			return 1;
 106	}
 107	return 0;
 108}
 109
 110/*
 111 * Add a memory region to the kernel e820 map.
 112 */
 113static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
 114					 int type)
 115{
 116	int x = e820x->nr_map;
 117
 118	if (x >= ARRAY_SIZE(e820x->map)) {
 119		printk(KERN_ERR "e820: too many entries; ignoring [mem %#010llx-%#010llx]\n",
 120		       (unsigned long long) start,
 121		       (unsigned long long) (start + size - 1));
 122		return;
 123	}
 124
 125	e820x->map[x].addr = start;
 126	e820x->map[x].size = size;
 127	e820x->map[x].type = type;
 128	e820x->nr_map++;
 129}
 130
 131void __init e820_add_region(u64 start, u64 size, int type)
 132{
 133	__e820_add_region(e820, start, size, type);
 134}
 135
 136static void __init e820_print_type(u32 type)
 137{
 138	switch (type) {
 139	case E820_RAM:
 140	case E820_RESERVED_KERN:
 141		printk(KERN_CONT "usable");
 142		break;
 143	case E820_RESERVED:
 144		printk(KERN_CONT "reserved");
 145		break;
 146	case E820_ACPI:
 147		printk(KERN_CONT "ACPI data");
 148		break;
 149	case E820_NVS:
 150		printk(KERN_CONT "ACPI NVS");
 151		break;
 152	case E820_UNUSABLE:
 153		printk(KERN_CONT "unusable");
 154		break;
 155	case E820_PMEM:
 156	case E820_PRAM:
 157		printk(KERN_CONT "persistent (type %u)", type);
 158		break;
 159	default:
 160		printk(KERN_CONT "type %u", type);
 161		break;
 162	}
 163}
 164
 165void __init e820_print_map(char *who)
 166{
 167	int i;
 168
 169	for (i = 0; i < e820->nr_map; i++) {
 170		printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who,
 171		       (unsigned long long) e820->map[i].addr,
 172		       (unsigned long long)
 173		       (e820->map[i].addr + e820->map[i].size - 1));
 174		e820_print_type(e820->map[i].type);
 175		printk(KERN_CONT "\n");
 176	}
 177}
 178
 179/*
 180 * Sanitize the BIOS e820 map.
 181 *
 182 * Some e820 responses include overlapping entries. The following
 183 * replaces the original e820 map with a new one, removing overlaps,
 184 * and resolving conflicting memory types in favor of highest
 185 * numbered type.
 186 *
 187 * The input parameter biosmap points to an array of 'struct
 188 * e820entry' which on entry has elements in the range [0, *pnr_map)
 189 * valid, and which has space for up to max_nr_map entries.
 190 * On return, the resulting sanitized e820 map entries will be in
 191 * overwritten in the same location, starting at biosmap.
 192 *
 193 * The integer pointed to by pnr_map must be valid on entry (the
 194 * current number of valid entries located at biosmap). If the
 195 * sanitizing succeeds the *pnr_map will be updated with the new
 196 * number of valid entries (something no more than max_nr_map).
 197 *
 198 * The return value from sanitize_e820_map() is zero if it
 199 * successfully 'sanitized' the map entries passed in, and is -1
 200 * if it did nothing, which can happen if either of (1) it was
 201 * only passed one map entry, or (2) any of the input map entries
 202 * were invalid (start + size < start, meaning that the size was
 203 * so big the described memory range wrapped around through zero.)
 204 *
 205 *	Visually we're performing the following
 206 *	(1,2,3,4 = memory types)...
 207 *
 208 *	Sample memory map (w/overlaps):
 209 *	   ____22__________________
 210 *	   ______________________4_
 211 *	   ____1111________________
 212 *	   _44_____________________
 213 *	   11111111________________
 214 *	   ____________________33__
 215 *	   ___________44___________
 216 *	   __________33333_________
 217 *	   ______________22________
 218 *	   ___________________2222_
 219 *	   _________111111111______
 220 *	   _____________________11_
 221 *	   _________________4______
 222 *
 223 *	Sanitized equivalent (no overlap):
 224 *	   1_______________________
 225 *	   _44_____________________
 226 *	   ___1____________________
 227 *	   ____22__________________
 228 *	   ______11________________
 229 *	   _________1______________
 230 *	   __________3_____________
 231 *	   ___________44___________
 232 *	   _____________33_________
 233 *	   _______________2________
 234 *	   ________________1_______
 235 *	   _________________4______
 236 *	   ___________________2____
 237 *	   ____________________33__
 238 *	   ______________________4_
 239 */
 240struct change_member {
 241	struct e820entry *pbios; /* pointer to original bios entry */
 242	unsigned long long addr; /* address for this change point */
 243};
 244
 245static int __init cpcompare(const void *a, const void *b)
 246{
 247	struct change_member * const *app = a, * const *bpp = b;
 248	const struct change_member *ap = *app, *bp = *bpp;
 249
 250	/*
 251	 * Inputs are pointers to two elements of change_point[].  If their
 252	 * addresses are unequal, their difference dominates.  If the addresses
 253	 * are equal, then consider one that represents the end of its region
 254	 * to be greater than one that does not.
 255	 */
 256	if (ap->addr != bp->addr)
 257		return ap->addr > bp->addr ? 1 : -1;
 258
 259	return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr);
 260}
 261
 262int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
 263			     u32 *pnr_map)
 264{
 
 
 
 
 265	static struct change_member change_point_list[2*E820_X_MAX] __initdata;
 266	static struct change_member *change_point[2*E820_X_MAX] __initdata;
 267	static struct e820entry *overlap_list[E820_X_MAX] __initdata;
 268	static struct e820entry new_bios[E820_X_MAX] __initdata;
 
 269	unsigned long current_type, last_type;
 270	unsigned long long last_addr;
 271	int chgidx;
 272	int overlap_entries;
 273	int new_bios_entry;
 274	int old_nr, new_nr, chg_nr;
 275	int i;
 276
 277	/* if there's only one memory region, don't bother */
 278	if (*pnr_map < 2)
 279		return -1;
 280
 281	old_nr = *pnr_map;
 282	BUG_ON(old_nr > max_nr_map);
 283
 284	/* bail out if we find any unreasonable addresses in bios map */
 285	for (i = 0; i < old_nr; i++)
 286		if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 287			return -1;
 288
 289	/* create pointers for initial change-point information (for sorting) */
 290	for (i = 0; i < 2 * old_nr; i++)
 291		change_point[i] = &change_point_list[i];
 292
 293	/* record all known change-points (starting and ending addresses),
 294	   omitting those that are for empty memory regions */
 295	chgidx = 0;
 296	for (i = 0; i < old_nr; i++)	{
 297		if (biosmap[i].size != 0) {
 298			change_point[chgidx]->addr = biosmap[i].addr;
 299			change_point[chgidx++]->pbios = &biosmap[i];
 300			change_point[chgidx]->addr = biosmap[i].addr +
 301				biosmap[i].size;
 302			change_point[chgidx++]->pbios = &biosmap[i];
 303		}
 304	}
 305	chg_nr = chgidx;
 306
 307	/* sort change-point list by memory addresses (low -> high) */
 308	sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 309
 310	/* create a new bios memory map, removing overlaps */
 311	overlap_entries = 0;	 /* number of entries in the overlap table */
 312	new_bios_entry = 0;	 /* index for creating new bios map entries */
 313	last_type = 0;		 /* start with undefined memory type */
 314	last_addr = 0;		 /* start with 0 as last starting address */
 315
 316	/* loop through change-points, determining affect on the new bios map */
 317	for (chgidx = 0; chgidx < chg_nr; chgidx++) {
 318		/* keep track of all overlapping bios entries */
 319		if (change_point[chgidx]->addr ==
 320		    change_point[chgidx]->pbios->addr) {
 321			/*
 322			 * add map entry to overlap list (> 1 entry
 323			 * implies an overlap)
 324			 */
 325			overlap_list[overlap_entries++] =
 326				change_point[chgidx]->pbios;
 327		} else {
 328			/*
 329			 * remove entry from list (order independent,
 330			 * so swap with last)
 331			 */
 332			for (i = 0; i < overlap_entries; i++) {
 333				if (overlap_list[i] ==
 334				    change_point[chgidx]->pbios)
 335					overlap_list[i] =
 336						overlap_list[overlap_entries-1];
 337			}
 338			overlap_entries--;
 339		}
 340		/*
 341		 * if there are overlapping entries, decide which
 342		 * "type" to use (larger value takes precedence --
 343		 * 1=usable, 2,3,4,4+=unusable)
 344		 */
 345		current_type = 0;
 346		for (i = 0; i < overlap_entries; i++)
 347			if (overlap_list[i]->type > current_type)
 348				current_type = overlap_list[i]->type;
 349		/*
 350		 * continue building up new bios map based on this
 351		 * information
 352		 */
 353		if (current_type != last_type || current_type == E820_PRAM) {
 354			if (last_type != 0)	 {
 355				new_bios[new_bios_entry].size =
 356					change_point[chgidx]->addr - last_addr;
 357				/*
 358				 * move forward only if the new size
 359				 * was non-zero
 360				 */
 361				if (new_bios[new_bios_entry].size != 0)
 362					/*
 363					 * no more space left for new
 364					 * bios entries ?
 365					 */
 366					if (++new_bios_entry >= max_nr_map)
 367						break;
 368			}
 369			if (current_type != 0)	{
 370				new_bios[new_bios_entry].addr =
 371					change_point[chgidx]->addr;
 372				new_bios[new_bios_entry].type = current_type;
 373				last_addr = change_point[chgidx]->addr;
 374			}
 375			last_type = current_type;
 376		}
 377	}
 378	/* retain count for new bios entries */
 379	new_nr = new_bios_entry;
 380
 381	/* copy new bios mapping into original location */
 382	memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
 383	*pnr_map = new_nr;
 384
 385	return 0;
 386}
 387
 388static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
 389{
 390	while (nr_map) {
 391		u64 start = biosmap->addr;
 392		u64 size = biosmap->size;
 393		u64 end = start + size - 1;
 394		u32 type = biosmap->type;
 395
 396		/* Overflow in 64 bits? Ignore the memory map. */
 397		if (start > end && likely(size))
 398			return -1;
 399
 400		e820_add_region(start, size, type);
 401
 402		biosmap++;
 403		nr_map--;
 404	}
 405	return 0;
 406}
 407
 408/*
 409 * Copy the BIOS e820 map into a safe place.
 410 *
 411 * Sanity-check it while we're at it..
 412 *
 413 * If we're lucky and live on a modern system, the setup code
 414 * will have given us a memory map that we can use to properly
 415 * set up memory.  If we aren't, we'll fake a memory map.
 416 */
 417static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
 418{
 419	/* Only one memory region (or negative)? Ignore it */
 420	if (nr_map < 2)
 421		return -1;
 422
 423	return __append_e820_map(biosmap, nr_map);
 424}
 425
 426static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
 427					u64 size, unsigned old_type,
 428					unsigned new_type)
 429{
 430	u64 end;
 431	unsigned int i;
 432	u64 real_updated_size = 0;
 433
 434	BUG_ON(old_type == new_type);
 435
 436	if (size > (ULLONG_MAX - start))
 437		size = ULLONG_MAX - start;
 438
 439	end = start + size;
 440	printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ",
 441	       (unsigned long long) start, (unsigned long long) (end - 1));
 
 442	e820_print_type(old_type);
 443	printk(KERN_CONT " ==> ");
 444	e820_print_type(new_type);
 445	printk(KERN_CONT "\n");
 446
 447	for (i = 0; i < e820x->nr_map; i++) {
 448		struct e820entry *ei = &e820x->map[i];
 449		u64 final_start, final_end;
 450		u64 ei_end;
 451
 452		if (ei->type != old_type)
 453			continue;
 454
 455		ei_end = ei->addr + ei->size;
 456		/* totally covered by new range? */
 457		if (ei->addr >= start && ei_end <= end) {
 458			ei->type = new_type;
 459			real_updated_size += ei->size;
 460			continue;
 461		}
 462
 463		/* new range is totally covered? */
 464		if (ei->addr < start && ei_end > end) {
 465			__e820_add_region(e820x, start, size, new_type);
 466			__e820_add_region(e820x, end, ei_end - end, ei->type);
 467			ei->size = start - ei->addr;
 468			real_updated_size += size;
 469			continue;
 470		}
 471
 472		/* partially covered */
 473		final_start = max(start, ei->addr);
 474		final_end = min(end, ei_end);
 475		if (final_start >= final_end)
 476			continue;
 477
 478		__e820_add_region(e820x, final_start, final_end - final_start,
 479				  new_type);
 480
 481		real_updated_size += final_end - final_start;
 482
 483		/*
 484		 * left range could be head or tail, so need to update
 485		 * size at first.
 486		 */
 487		ei->size -= final_end - final_start;
 488		if (ei->addr < final_start)
 489			continue;
 490		ei->addr = final_end;
 491	}
 492	return real_updated_size;
 493}
 494
 495u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
 496			     unsigned new_type)
 497{
 498	return __e820_update_range(e820, start, size, old_type, new_type);
 499}
 500
 501static u64 __init e820_update_range_saved(u64 start, u64 size,
 502					  unsigned old_type, unsigned new_type)
 503{
 504	return __e820_update_range(e820_saved, start, size, old_type,
 505				     new_type);
 506}
 507
 508/* make e820 not cover the range */
 509u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
 510			     int checktype)
 511{
 512	int i;
 513	u64 end;
 514	u64 real_removed_size = 0;
 515
 516	if (size > (ULLONG_MAX - start))
 517		size = ULLONG_MAX - start;
 518
 519	end = start + size;
 520	printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ",
 521	       (unsigned long long) start, (unsigned long long) (end - 1));
 
 522	if (checktype)
 523		e820_print_type(old_type);
 524	printk(KERN_CONT "\n");
 525
 526	for (i = 0; i < e820->nr_map; i++) {
 527		struct e820entry *ei = &e820->map[i];
 528		u64 final_start, final_end;
 529		u64 ei_end;
 530
 531		if (checktype && ei->type != old_type)
 532			continue;
 533
 534		ei_end = ei->addr + ei->size;
 535		/* totally covered? */
 536		if (ei->addr >= start && ei_end <= end) {
 537			real_removed_size += ei->size;
 538			memset(ei, 0, sizeof(struct e820entry));
 539			continue;
 540		}
 541
 542		/* new range is totally covered? */
 543		if (ei->addr < start && ei_end > end) {
 544			e820_add_region(end, ei_end - end, ei->type);
 545			ei->size = start - ei->addr;
 546			real_removed_size += size;
 547			continue;
 548		}
 549
 550		/* partially covered */
 551		final_start = max(start, ei->addr);
 552		final_end = min(end, ei_end);
 553		if (final_start >= final_end)
 554			continue;
 555		real_removed_size += final_end - final_start;
 556
 557		/*
 558		 * left range could be head or tail, so need to update
 559		 * size at first.
 560		 */
 561		ei->size -= final_end - final_start;
 562		if (ei->addr < final_start)
 563			continue;
 564		ei->addr = final_end;
 565	}
 566	return real_removed_size;
 567}
 568
 569void __init update_e820(void)
 570{
 571	if (sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map))
 
 
 
 572		return;
 573	printk(KERN_INFO "e820: modified physical RAM map:\n");
 
 574	e820_print_map("modified");
 575}
 576static void __init update_e820_saved(void)
 577{
 578	sanitize_e820_map(e820_saved->map, ARRAY_SIZE(e820_saved->map),
 579				&e820_saved->nr_map);
 
 
 
 
 580}
 581#define MAX_GAP_END 0x100000000ull
 582/*
 583 * Search for a gap in the e820 memory space from start_addr to end_addr.
 584 */
 585__init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
 586		unsigned long start_addr, unsigned long long end_addr)
 587{
 588	unsigned long long last;
 589	int i = e820->nr_map;
 590	int found = 0;
 591
 592	last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
 593
 594	while (--i >= 0) {
 595		unsigned long long start = e820->map[i].addr;
 596		unsigned long long end = start + e820->map[i].size;
 597
 598		if (end < start_addr)
 599			continue;
 600
 601		/*
 602		 * Since "last" is at most 4GB, we know we'll
 603		 * fit in 32 bits if this condition is true
 604		 */
 605		if (last > end) {
 606			unsigned long gap = last - end;
 607
 608			if (gap >= *gapsize) {
 609				*gapsize = gap;
 610				*gapstart = end;
 611				found = 1;
 612			}
 613		}
 614		if (start < last)
 615			last = start;
 616	}
 617	return found;
 618}
 619
 620/*
 621 * Search for the biggest gap in the low 32 bits of the e820
 622 * memory space.  We pass this space to PCI to assign MMIO resources
 623 * for hotplug or unconfigured devices in.
 624 * Hopefully the BIOS let enough space left.
 625 */
 626__init void e820_setup_gap(void)
 627{
 628	unsigned long gapstart, gapsize;
 629	int found;
 630
 631	gapstart = 0x10000000;
 632	gapsize = 0x400000;
 633	found  = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
 634
 635#ifdef CONFIG_X86_64
 636	if (!found) {
 637		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
 638		printk(KERN_ERR
 639	"e820: cannot find a gap in the 32bit address range\n"
 640	"e820: PCI devices with unassigned 32bit BARs may break!\n");
 641	}
 642#endif
 643
 644	/*
 645	 * e820_reserve_resources_late protect stolen RAM already
 646	 */
 647	pci_mem_start = gapstart;
 648
 649	printk(KERN_INFO
 650	       "e820: [mem %#010lx-%#010lx] available for PCI devices\n",
 651	       gapstart, gapstart + gapsize - 1);
 652}
 653
 654/*
 655 * Called late during init, in free_initmem().
 656 *
 657 * Initial e820 and e820_saved are largish __initdata arrays.
 658 * Copy them to (usually much smaller) dynamically allocated area.
 659 * This is done after all tweaks we ever do to them:
 660 * all functions which modify them are __init functions,
 661 * they won't exist after this point.
 662 */
 663__init void e820_reallocate_tables(void)
 664{
 665	struct e820map *n;
 666	int size;
 667
 668	size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820->nr_map;
 669	n = kmalloc(size, GFP_KERNEL);
 670	BUG_ON(!n);
 671	memcpy(n, e820, size);
 672	e820 = n;
 673
 674	size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820_saved->nr_map;
 675	n = kmalloc(size, GFP_KERNEL);
 676	BUG_ON(!n);
 677	memcpy(n, e820_saved, size);
 678	e820_saved = n;
 679}
 680
 681/**
 682 * Because of the size limitation of struct boot_params, only first
 683 * 128 E820 memory entries are passed to kernel via
 684 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
 685 * linked list of struct setup_data, which is parsed here.
 686 */
 687void __init parse_e820_ext(u64 phys_addr, u32 data_len)
 688{
 689	int entries;
 690	struct e820entry *extmap;
 691	struct setup_data *sdata;
 692
 693	sdata = early_memremap(phys_addr, data_len);
 694	entries = sdata->len / sizeof(struct e820entry);
 695	extmap = (struct e820entry *)(sdata->data);
 696	__append_e820_map(extmap, entries);
 697	sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map);
 698	early_memunmap(sdata, data_len);
 699	printk(KERN_INFO "e820: extended physical RAM map:\n");
 700	e820_print_map("extended");
 701}
 702
 703#if defined(CONFIG_X86_64) || \
 704	(defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
 705/**
 706 * Find the ranges of physical addresses that do not correspond to
 707 * e820 RAM areas and mark the corresponding pages as nosave for
 708 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
 709 *
 710 * This function requires the e820 map to be sorted and without any
 711 * overlapping entries.
 712 */
 713void __init e820_mark_nosave_regions(unsigned long limit_pfn)
 714{
 715	int i;
 716	unsigned long pfn = 0;
 717
 718	for (i = 0; i < e820->nr_map; i++) {
 719		struct e820entry *ei = &e820->map[i];
 
 720
 721		if (pfn < PFN_UP(ei->addr))
 722			register_nosave_region(pfn, PFN_UP(ei->addr));
 723
 724		pfn = PFN_DOWN(ei->addr + ei->size);
 725
 726		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
 727			register_nosave_region(PFN_UP(ei->addr), pfn);
 728
 729		if (pfn >= limit_pfn)
 730			break;
 731	}
 732}
 733#endif
 734
 735#ifdef CONFIG_ACPI
 736/**
 737 * Mark ACPI NVS memory region, so that we can save/restore it during
 738 * hibernation and the subsequent resume.
 739 */
 740static int __init e820_mark_nvs_memory(void)
 741{
 742	int i;
 743
 744	for (i = 0; i < e820->nr_map; i++) {
 745		struct e820entry *ei = &e820->map[i];
 746
 747		if (ei->type == E820_NVS)
 748			acpi_nvs_register(ei->addr, ei->size);
 749	}
 750
 751	return 0;
 752}
 753core_initcall(e820_mark_nvs_memory);
 754#endif
 755
 756/*
 757 * pre allocated 4k and reserved it in memblock and e820_saved
 758 */
 759u64 __init early_reserve_e820(u64 size, u64 align)
 760{
 
 761	u64 addr;
 
 762
 763	addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
 764	if (addr) {
 765		e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED);
 766		printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n");
 767		update_e820_saved();
 
 768	}
 769
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 770	return addr;
 771}
 772
 773#ifdef CONFIG_X86_32
 774# ifdef CONFIG_X86_PAE
 775#  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
 776# else
 777#  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
 778# endif
 779#else /* CONFIG_X86_32 */
 780# define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
 781#endif
 782
 783/*
 784 * Find the highest page frame number we have available
 785 */
 786static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
 787{
 788	int i;
 789	unsigned long last_pfn = 0;
 790	unsigned long max_arch_pfn = MAX_ARCH_PFN;
 791
 792	for (i = 0; i < e820->nr_map; i++) {
 793		struct e820entry *ei = &e820->map[i];
 794		unsigned long start_pfn;
 795		unsigned long end_pfn;
 796
 797		if (ei->type != type)
 798			continue;
 799
 800		start_pfn = ei->addr >> PAGE_SHIFT;
 801		end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
 802
 803		if (start_pfn >= limit_pfn)
 804			continue;
 805		if (end_pfn > limit_pfn) {
 806			last_pfn = limit_pfn;
 807			break;
 808		}
 809		if (end_pfn > last_pfn)
 810			last_pfn = end_pfn;
 811	}
 812
 813	if (last_pfn > max_arch_pfn)
 814		last_pfn = max_arch_pfn;
 815
 816	printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
 817			 last_pfn, max_arch_pfn);
 818	return last_pfn;
 819}
 820unsigned long __init e820_end_of_ram_pfn(void)
 821{
 822	return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
 823}
 824
 825unsigned long __init e820_end_of_low_ram_pfn(void)
 826{
 827	return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_RAM);
 828}
 829
 830static void __init early_panic(char *msg)
 831{
 832	early_printk(msg);
 833	panic(msg);
 834}
 835
 836static int userdef __initdata;
 837
 838/* "mem=nopentium" disables the 4MB page tables. */
 839static int __init parse_memopt(char *p)
 840{
 841	u64 mem_size;
 842
 843	if (!p)
 844		return -EINVAL;
 845
 846	if (!strcmp(p, "nopentium")) {
 847#ifdef CONFIG_X86_32
 848		setup_clear_cpu_cap(X86_FEATURE_PSE);
 849		return 0;
 850#else
 851		printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
 852		return -EINVAL;
 853#endif
 854	}
 855
 856	userdef = 1;
 857	mem_size = memparse(p, &p);
 858	/* don't remove all of memory when handling "mem={invalid}" param */
 859	if (mem_size == 0)
 860		return -EINVAL;
 861	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
 862
 863	return 0;
 864}
 865early_param("mem", parse_memopt);
 866
 867static int __init parse_memmap_one(char *p)
 868{
 869	char *oldp;
 870	u64 start_at, mem_size;
 871
 872	if (!p)
 873		return -EINVAL;
 874
 875	if (!strncmp(p, "exactmap", 8)) {
 876#ifdef CONFIG_CRASH_DUMP
 877		/*
 878		 * If we are doing a crash dump, we still need to know
 879		 * the real mem size before original memory map is
 880		 * reset.
 881		 */
 882		saved_max_pfn = e820_end_of_ram_pfn();
 883#endif
 884		e820->nr_map = 0;
 885		userdef = 1;
 886		return 0;
 887	}
 888
 889	oldp = p;
 890	mem_size = memparse(p, &p);
 891	if (p == oldp)
 892		return -EINVAL;
 893
 894	userdef = 1;
 895	if (*p == '@') {
 896		start_at = memparse(p+1, &p);
 897		e820_add_region(start_at, mem_size, E820_RAM);
 898	} else if (*p == '#') {
 899		start_at = memparse(p+1, &p);
 900		e820_add_region(start_at, mem_size, E820_ACPI);
 901	} else if (*p == '$') {
 902		start_at = memparse(p+1, &p);
 903		e820_add_region(start_at, mem_size, E820_RESERVED);
 904	} else if (*p == '!') {
 905		start_at = memparse(p+1, &p);
 906		e820_add_region(start_at, mem_size, E820_PRAM);
 907	} else
 908		e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
 909
 910	return *p == '\0' ? 0 : -EINVAL;
 911}
 912static int __init parse_memmap_opt(char *str)
 913{
 914	while (str) {
 915		char *k = strchr(str, ',');
 916
 917		if (k)
 918			*k++ = 0;
 919
 920		parse_memmap_one(str);
 921		str = k;
 922	}
 923
 924	return 0;
 925}
 926early_param("memmap", parse_memmap_opt);
 927
 928void __init finish_e820_parsing(void)
 929{
 930	if (userdef) {
 931		if (sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map),
 932					&e820->nr_map) < 0)
 
 933			early_panic("Invalid user supplied memory map");
 
 934
 935		printk(KERN_INFO "e820: user-defined physical RAM map:\n");
 936		e820_print_map("user");
 937	}
 938}
 939
 940static const char *__init e820_type_to_string(int e820_type)
 941{
 942	switch (e820_type) {
 943	case E820_RESERVED_KERN:
 944	case E820_RAM:	return "System RAM";
 945	case E820_ACPI:	return "ACPI Tables";
 946	case E820_NVS:	return "ACPI Non-volatile Storage";
 947	case E820_UNUSABLE:	return "Unusable memory";
 948	case E820_PRAM: return "Persistent Memory (legacy)";
 949	case E820_PMEM: return "Persistent Memory";
 950	default:	return "reserved";
 951	}
 952}
 953
 954static unsigned long __init e820_type_to_iomem_type(int e820_type)
 955{
 956	switch (e820_type) {
 957	case E820_RESERVED_KERN:
 958	case E820_RAM:
 959		return IORESOURCE_SYSTEM_RAM;
 960	case E820_ACPI:
 961	case E820_NVS:
 962	case E820_UNUSABLE:
 963	case E820_PRAM:
 964	case E820_PMEM:
 965	default:
 966		return IORESOURCE_MEM;
 967	}
 968}
 969
 970static unsigned long __init e820_type_to_iores_desc(int e820_type)
 971{
 972	switch (e820_type) {
 973	case E820_ACPI:
 974		return IORES_DESC_ACPI_TABLES;
 975	case E820_NVS:
 976		return IORES_DESC_ACPI_NV_STORAGE;
 977	case E820_PMEM:
 978		return IORES_DESC_PERSISTENT_MEMORY;
 979	case E820_PRAM:
 980		return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
 981	case E820_RESERVED_KERN:
 982	case E820_RAM:
 983	case E820_UNUSABLE:
 984	default:
 985		return IORES_DESC_NONE;
 986	}
 987}
 988
 989static bool __init do_mark_busy(u32 type, struct resource *res)
 990{
 991	/* this is the legacy bios/dos rom-shadow + mmio region */
 992	if (res->start < (1ULL<<20))
 993		return true;
 994
 995	/*
 996	 * Treat persistent memory like device memory, i.e. reserve it
 997	 * for exclusive use of a driver
 998	 */
 999	switch (type) {
1000	case E820_RESERVED:
1001	case E820_PRAM:
1002	case E820_PMEM:
1003		return false;
1004	default:
1005		return true;
1006	}
1007}
1008
1009/*
1010 * Mark e820 reserved areas as busy for the resource manager.
1011 */
1012static struct resource __initdata *e820_res;
1013void __init e820_reserve_resources(void)
1014{
1015	int i;
1016	struct resource *res;
1017	u64 end;
1018
1019	res = alloc_bootmem(sizeof(struct resource) * e820->nr_map);
1020	e820_res = res;
1021	for (i = 0; i < e820->nr_map; i++) {
1022		end = e820->map[i].addr + e820->map[i].size - 1;
1023		if (end != (resource_size_t)end) {
1024			res++;
1025			continue;
1026		}
1027		res->name = e820_type_to_string(e820->map[i].type);
1028		res->start = e820->map[i].addr;
1029		res->end = end;
1030
1031		res->flags = e820_type_to_iomem_type(e820->map[i].type);
1032		res->desc = e820_type_to_iores_desc(e820->map[i].type);
1033
1034		/*
1035		 * don't register the region that could be conflicted with
1036		 * pci device BAR resource and insert them later in
1037		 * pcibios_resource_survey()
1038		 */
1039		if (do_mark_busy(e820->map[i].type, res)) {
1040			res->flags |= IORESOURCE_BUSY;
1041			insert_resource(&iomem_resource, res);
1042		}
1043		res++;
1044	}
1045
1046	for (i = 0; i < e820_saved->nr_map; i++) {
1047		struct e820entry *entry = &e820_saved->map[i];
1048		firmware_map_add_early(entry->addr,
1049			entry->addr + entry->size,
1050			e820_type_to_string(entry->type));
1051	}
1052}
1053
1054/* How much should we pad RAM ending depending on where it is? */
1055static unsigned long __init ram_alignment(resource_size_t pos)
1056{
1057	unsigned long mb = pos >> 20;
1058
1059	/* To 64kB in the first megabyte */
1060	if (!mb)
1061		return 64*1024;
1062
1063	/* To 1MB in the first 16MB */
1064	if (mb < 16)
1065		return 1024*1024;
1066
1067	/* To 64MB for anything above that */
1068	return 64*1024*1024;
1069}
1070
1071#define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1072
1073void __init e820_reserve_resources_late(void)
1074{
1075	int i;
1076	struct resource *res;
1077
1078	res = e820_res;
1079	for (i = 0; i < e820->nr_map; i++) {
1080		if (!res->parent && res->end)
1081			insert_resource_expand_to_fit(&iomem_resource, res);
1082		res++;
1083	}
1084
1085	/*
1086	 * Try to bump up RAM regions to reasonable boundaries to
1087	 * avoid stolen RAM:
1088	 */
1089	for (i = 0; i < e820->nr_map; i++) {
1090		struct e820entry *entry = &e820->map[i];
1091		u64 start, end;
1092
1093		if (entry->type != E820_RAM)
1094			continue;
1095		start = entry->addr + entry->size;
1096		end = round_up(start, ram_alignment(start)) - 1;
1097		if (end > MAX_RESOURCE_SIZE)
1098			end = MAX_RESOURCE_SIZE;
1099		if (start >= end)
1100			continue;
1101		printk(KERN_DEBUG
1102		       "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n",
1103		       start, end);
1104		reserve_region_with_split(&iomem_resource, start, end,
1105					  "RAM buffer");
1106	}
1107}
1108
1109char *__init default_machine_specific_memory_setup(void)
1110{
1111	char *who = "BIOS-e820";
1112	u32 new_nr;
1113	/*
1114	 * Try to copy the BIOS-supplied E820-map.
1115	 *
1116	 * Otherwise fake a memory map; one section from 0k->640k,
1117	 * the next section from 1mb->appropriate_mem_k
1118	 */
1119	new_nr = boot_params.e820_entries;
1120	sanitize_e820_map(boot_params.e820_map,
1121			ARRAY_SIZE(boot_params.e820_map),
1122			&new_nr);
1123	boot_params.e820_entries = new_nr;
1124	if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1125	  < 0) {
1126		u64 mem_size;
1127
1128		/* compare results from other methods and take the greater */
1129		if (boot_params.alt_mem_k
1130		    < boot_params.screen_info.ext_mem_k) {
1131			mem_size = boot_params.screen_info.ext_mem_k;
1132			who = "BIOS-88";
1133		} else {
1134			mem_size = boot_params.alt_mem_k;
1135			who = "BIOS-e801";
1136		}
1137
1138		e820->nr_map = 0;
1139		e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1140		e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1141	}
1142
1143	/* In case someone cares... */
1144	return who;
1145}
1146
1147void __init setup_memory_map(void)
1148{
1149	char *who;
1150
1151	who = x86_init.resources.memory_setup();
1152	memcpy(e820_saved, e820, sizeof(struct e820map));
1153	printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n");
1154	e820_print_map(who);
1155}
1156
1157void __init memblock_x86_fill(void)
1158{
1159	int i;
1160	u64 end;
1161
1162	/*
1163	 * EFI may have more than 128 entries
1164	 * We are safe to enable resizing, beause memblock_x86_fill()
1165	 * is rather later for x86
1166	 */
1167	memblock_allow_resize();
1168
1169	for (i = 0; i < e820->nr_map; i++) {
1170		struct e820entry *ei = &e820->map[i];
1171
1172		end = ei->addr + ei->size;
1173		if (end != (resource_size_t)end)
1174			continue;
1175
1176		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1177			continue;
1178
1179		memblock_add(ei->addr, ei->size);
1180	}
1181
1182	/* throw away partial pages */
1183	memblock_trim_memory(PAGE_SIZE);
1184
1185	memblock_dump_all();
1186}
1187
1188void __init memblock_find_dma_reserve(void)
1189{
1190#ifdef CONFIG_X86_64
1191	u64 nr_pages = 0, nr_free_pages = 0;
1192	unsigned long start_pfn, end_pfn;
1193	phys_addr_t start, end;
1194	int i;
1195	u64 u;
1196
1197	/*
1198	 * need to find out used area below MAX_DMA_PFN
1199	 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1200	 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1201	 */
1202	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1203		start_pfn = min(start_pfn, MAX_DMA_PFN);
1204		end_pfn = min(end_pfn, MAX_DMA_PFN);
1205		nr_pages += end_pfn - start_pfn;
1206	}
1207
1208	for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1209				NULL) {
1210		start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1211		end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1212		if (start_pfn < end_pfn)
1213			nr_free_pages += end_pfn - start_pfn;
1214	}
1215
1216	set_dma_reserve(nr_pages - nr_free_pages);
1217#endif
1218}