xref: /linux/arch/x86/kernel/e820.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 
28 /*
29  * The e820 map is the map that gets modified e.g. with command line parameters
30  * and that is also registered with modifications in the kernel resource tree
31  * with the iomem_resource as parent.
32  *
33  * The e820_saved is directly saved after the BIOS-provided memory map is
34  * copied. It doesn't get modified afterwards. It's registered for the
35  * /sys/firmware/memmap interface.
36  *
37  * That memory map is not modified and is used as base for kexec. The kexec'd
38  * kernel should get the same memory map as the firmware provides. Then the
39  * user can e.g. boot the original kernel with mem=1G while still booting the
40  * next kernel with full memory.
41  */
42 struct e820map e820;
43 struct e820map e820_saved;
44 
45 /* For PCI or other memory-mapped resources */
46 unsigned long pci_mem_start = 0xaeedbabe;
47 #ifdef CONFIG_PCI
48 EXPORT_SYMBOL(pci_mem_start);
49 #endif
50 
51 /*
52  * This function checks if any part of the range <start,end> is mapped
53  * with type.
54  */
55 int
56 e820_any_mapped(u64 start, u64 end, unsigned type)
57 {
58 	int i;
59 
60 	for (i = 0; i < e820.nr_map; i++) {
61 		struct e820entry *ei = &e820.map[i];
62 
63 		if (type && ei->type != type)
64 			continue;
65 		if (ei->addr >= end || ei->addr + ei->size <= start)
66 			continue;
67 		return 1;
68 	}
69 	return 0;
70 }
71 EXPORT_SYMBOL_GPL(e820_any_mapped);
72 
73 /*
74  * This function checks if the entire range <start,end> is mapped with type.
75  *
76  * Note: this function only works correct if the e820 table is sorted and
77  * not-overlapping, which is the case
78  */
79 int __init e820_all_mapped(u64 start, u64 end, unsigned type)
80 {
81 	int i;
82 
83 	for (i = 0; i < e820.nr_map; i++) {
84 		struct e820entry *ei = &e820.map[i];
85 
86 		if (type && ei->type != type)
87 			continue;
88 		/* is the region (part) in overlap with the current region ?*/
89 		if (ei->addr >= end || ei->addr + ei->size <= start)
90 			continue;
91 
92 		/* if the region is at the beginning of <start,end> we move
93 		 * start to the end of the region since it's ok until there
94 		 */
95 		if (ei->addr <= start)
96 			start = ei->addr + ei->size;
97 		/*
98 		 * if start is now at or beyond end, we're done, full
99 		 * coverage
100 		 */
101 		if (start >= end)
102 			return 1;
103 	}
104 	return 0;
105 }
106 
107 /*
108  * Add a memory region to the kernel e820 map.
109  */
110 static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
111 					 int type)
112 {
113 	int x = e820x->nr_map;
114 
115 	if (x >= ARRAY_SIZE(e820x->map)) {
116 		printk(KERN_ERR "e820: too many entries; ignoring [mem %#010llx-%#010llx]\n",
117 		       (unsigned long long) start,
118 		       (unsigned long long) (start + size - 1));
119 		return;
120 	}
121 
122 	e820x->map[x].addr = start;
123 	e820x->map[x].size = size;
124 	e820x->map[x].type = type;
125 	e820x->nr_map++;
126 }
127 
128 void __init e820_add_region(u64 start, u64 size, int type)
129 {
130 	__e820_add_region(&e820, start, size, type);
131 }
132 
133 static void __init e820_print_type(u32 type)
134 {
135 	switch (type) {
136 	case E820_RAM:
137 	case E820_RESERVED_KERN:
138 		printk(KERN_CONT "usable");
139 		break;
140 	case E820_RESERVED:
141 		printk(KERN_CONT "reserved");
142 		break;
143 	case E820_ACPI:
144 		printk(KERN_CONT "ACPI data");
145 		break;
146 	case E820_NVS:
147 		printk(KERN_CONT "ACPI NVS");
148 		break;
149 	case E820_UNUSABLE:
150 		printk(KERN_CONT "unusable");
151 		break;
152 	case E820_PMEM:
153 	case E820_PRAM:
154 		printk(KERN_CONT "persistent (type %u)", type);
155 		break;
156 	default:
157 		printk(KERN_CONT "type %u", type);
158 		break;
159 	}
160 }
161 
162 void __init e820_print_map(char *who)
163 {
164 	int i;
165 
166 	for (i = 0; i < e820.nr_map; i++) {
167 		printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who,
168 		       (unsigned long long) e820.map[i].addr,
169 		       (unsigned long long)
170 		       (e820.map[i].addr + e820.map[i].size - 1));
171 		e820_print_type(e820.map[i].type);
172 		printk(KERN_CONT "\n");
173 	}
174 }
175 
176 /*
177  * Sanitize the BIOS e820 map.
178  *
179  * Some e820 responses include overlapping entries. The following
180  * replaces the original e820 map with a new one, removing overlaps,
181  * and resolving conflicting memory types in favor of highest
182  * numbered type.
183  *
184  * The input parameter biosmap points to an array of 'struct
185  * e820entry' which on entry has elements in the range [0, *pnr_map)
186  * valid, and which has space for up to max_nr_map entries.
187  * On return, the resulting sanitized e820 map entries will be in
188  * overwritten in the same location, starting at biosmap.
189  *
190  * The integer pointed to by pnr_map must be valid on entry (the
191  * current number of valid entries located at biosmap). If the
192  * sanitizing succeeds the *pnr_map will be updated with the new
193  * number of valid entries (something no more than max_nr_map).
194  *
195  * The return value from sanitize_e820_map() is zero if it
196  * successfully 'sanitized' the map entries passed in, and is -1
197  * if it did nothing, which can happen if either of (1) it was
198  * only passed one map entry, or (2) any of the input map entries
199  * were invalid (start + size < start, meaning that the size was
200  * so big the described memory range wrapped around through zero.)
201  *
202  *	Visually we're performing the following
203  *	(1,2,3,4 = memory types)...
204  *
205  *	Sample memory map (w/overlaps):
206  *	   ____22__________________
207  *	   ______________________4_
208  *	   ____1111________________
209  *	   _44_____________________
210  *	   11111111________________
211  *	   ____________________33__
212  *	   ___________44___________
213  *	   __________33333_________
214  *	   ______________22________
215  *	   ___________________2222_
216  *	   _________111111111______
217  *	   _____________________11_
218  *	   _________________4______
219  *
220  *	Sanitized equivalent (no overlap):
221  *	   1_______________________
222  *	   _44_____________________
223  *	   ___1____________________
224  *	   ____22__________________
225  *	   ______11________________
226  *	   _________1______________
227  *	   __________3_____________
228  *	   ___________44___________
229  *	   _____________33_________
230  *	   _______________2________
231  *	   ________________1_______
232  *	   _________________4______
233  *	   ___________________2____
234  *	   ____________________33__
235  *	   ______________________4_
236  */
237 struct change_member {
238 	struct e820entry *pbios; /* pointer to original bios entry */
239 	unsigned long long addr; /* address for this change point */
240 };
241 
242 static int __init cpcompare(const void *a, const void *b)
243 {
244 	struct change_member * const *app = a, * const *bpp = b;
245 	const struct change_member *ap = *app, *bp = *bpp;
246 
247 	/*
248 	 * Inputs are pointers to two elements of change_point[].  If their
249 	 * addresses are unequal, their difference dominates.  If the addresses
250 	 * are equal, then consider one that represents the end of its region
251 	 * to be greater than one that does not.
252 	 */
253 	if (ap->addr != bp->addr)
254 		return ap->addr > bp->addr ? 1 : -1;
255 
256 	return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr);
257 }
258 
259 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
260 			     u32 *pnr_map)
261 {
262 	static struct change_member change_point_list[2*E820_X_MAX] __initdata;
263 	static struct change_member *change_point[2*E820_X_MAX] __initdata;
264 	static struct e820entry *overlap_list[E820_X_MAX] __initdata;
265 	static struct e820entry new_bios[E820_X_MAX] __initdata;
266 	unsigned long current_type, last_type;
267 	unsigned long long last_addr;
268 	int chgidx;
269 	int overlap_entries;
270 	int new_bios_entry;
271 	int old_nr, new_nr, chg_nr;
272 	int i;
273 
274 	/* if there's only one memory region, don't bother */
275 	if (*pnr_map < 2)
276 		return -1;
277 
278 	old_nr = *pnr_map;
279 	BUG_ON(old_nr > max_nr_map);
280 
281 	/* bail out if we find any unreasonable addresses in bios map */
282 	for (i = 0; i < old_nr; i++)
283 		if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
284 			return -1;
285 
286 	/* create pointers for initial change-point information (for sorting) */
287 	for (i = 0; i < 2 * old_nr; i++)
288 		change_point[i] = &change_point_list[i];
289 
290 	/* record all known change-points (starting and ending addresses),
291 	   omitting those that are for empty memory regions */
292 	chgidx = 0;
293 	for (i = 0; i < old_nr; i++)	{
294 		if (biosmap[i].size != 0) {
295 			change_point[chgidx]->addr = biosmap[i].addr;
296 			change_point[chgidx++]->pbios = &biosmap[i];
297 			change_point[chgidx]->addr = biosmap[i].addr +
298 				biosmap[i].size;
299 			change_point[chgidx++]->pbios = &biosmap[i];
300 		}
301 	}
302 	chg_nr = chgidx;
303 
304 	/* sort change-point list by memory addresses (low -> high) */
305 	sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL);
306 
307 	/* create a new bios memory map, removing overlaps */
308 	overlap_entries = 0;	 /* number of entries in the overlap table */
309 	new_bios_entry = 0;	 /* index for creating new bios map entries */
310 	last_type = 0;		 /* start with undefined memory type */
311 	last_addr = 0;		 /* start with 0 as last starting address */
312 
313 	/* loop through change-points, determining affect on the new bios map */
314 	for (chgidx = 0; chgidx < chg_nr; chgidx++) {
315 		/* keep track of all overlapping bios entries */
316 		if (change_point[chgidx]->addr ==
317 		    change_point[chgidx]->pbios->addr) {
318 			/*
319 			 * add map entry to overlap list (> 1 entry
320 			 * implies an overlap)
321 			 */
322 			overlap_list[overlap_entries++] =
323 				change_point[chgidx]->pbios;
324 		} else {
325 			/*
326 			 * remove entry from list (order independent,
327 			 * so swap with last)
328 			 */
329 			for (i = 0; i < overlap_entries; i++) {
330 				if (overlap_list[i] ==
331 				    change_point[chgidx]->pbios)
332 					overlap_list[i] =
333 						overlap_list[overlap_entries-1];
334 			}
335 			overlap_entries--;
336 		}
337 		/*
338 		 * if there are overlapping entries, decide which
339 		 * "type" to use (larger value takes precedence --
340 		 * 1=usable, 2,3,4,4+=unusable)
341 		 */
342 		current_type = 0;
343 		for (i = 0; i < overlap_entries; i++)
344 			if (overlap_list[i]->type > current_type)
345 				current_type = overlap_list[i]->type;
346 		/*
347 		 * continue building up new bios map based on this
348 		 * information
349 		 */
350 		if (current_type != last_type || current_type == E820_PRAM) {
351 			if (last_type != 0)	 {
352 				new_bios[new_bios_entry].size =
353 					change_point[chgidx]->addr - last_addr;
354 				/*
355 				 * move forward only if the new size
356 				 * was non-zero
357 				 */
358 				if (new_bios[new_bios_entry].size != 0)
359 					/*
360 					 * no more space left for new
361 					 * bios entries ?
362 					 */
363 					if (++new_bios_entry >= max_nr_map)
364 						break;
365 			}
366 			if (current_type != 0)	{
367 				new_bios[new_bios_entry].addr =
368 					change_point[chgidx]->addr;
369 				new_bios[new_bios_entry].type = current_type;
370 				last_addr = change_point[chgidx]->addr;
371 			}
372 			last_type = current_type;
373 		}
374 	}
375 	/* retain count for new bios entries */
376 	new_nr = new_bios_entry;
377 
378 	/* copy new bios mapping into original location */
379 	memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
380 	*pnr_map = new_nr;
381 
382 	return 0;
383 }
384 
385 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
386 {
387 	while (nr_map) {
388 		u64 start = biosmap->addr;
389 		u64 size = biosmap->size;
390 		u64 end = start + size;
391 		u32 type = biosmap->type;
392 
393 		/* Overflow in 64 bits? Ignore the memory map. */
394 		if (start > end)
395 			return -1;
396 
397 		e820_add_region(start, size, type);
398 
399 		biosmap++;
400 		nr_map--;
401 	}
402 	return 0;
403 }
404 
405 /*
406  * Copy the BIOS e820 map into a safe place.
407  *
408  * Sanity-check it while we're at it..
409  *
410  * If we're lucky and live on a modern system, the setup code
411  * will have given us a memory map that we can use to properly
412  * set up memory.  If we aren't, we'll fake a memory map.
413  */
414 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
415 {
416 	/* Only one memory region (or negative)? Ignore it */
417 	if (nr_map < 2)
418 		return -1;
419 
420 	return __append_e820_map(biosmap, nr_map);
421 }
422 
423 static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
424 					u64 size, unsigned old_type,
425 					unsigned new_type)
426 {
427 	u64 end;
428 	unsigned int i;
429 	u64 real_updated_size = 0;
430 
431 	BUG_ON(old_type == new_type);
432 
433 	if (size > (ULLONG_MAX - start))
434 		size = ULLONG_MAX - start;
435 
436 	end = start + size;
437 	printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ",
438 	       (unsigned long long) start, (unsigned long long) (end - 1));
439 	e820_print_type(old_type);
440 	printk(KERN_CONT " ==> ");
441 	e820_print_type(new_type);
442 	printk(KERN_CONT "\n");
443 
444 	for (i = 0; i < e820x->nr_map; i++) {
445 		struct e820entry *ei = &e820x->map[i];
446 		u64 final_start, final_end;
447 		u64 ei_end;
448 
449 		if (ei->type != old_type)
450 			continue;
451 
452 		ei_end = ei->addr + ei->size;
453 		/* totally covered by new range? */
454 		if (ei->addr >= start && ei_end <= end) {
455 			ei->type = new_type;
456 			real_updated_size += ei->size;
457 			continue;
458 		}
459 
460 		/* new range is totally covered? */
461 		if (ei->addr < start && ei_end > end) {
462 			__e820_add_region(e820x, start, size, new_type);
463 			__e820_add_region(e820x, end, ei_end - end, ei->type);
464 			ei->size = start - ei->addr;
465 			real_updated_size += size;
466 			continue;
467 		}
468 
469 		/* partially covered */
470 		final_start = max(start, ei->addr);
471 		final_end = min(end, ei_end);
472 		if (final_start >= final_end)
473 			continue;
474 
475 		__e820_add_region(e820x, final_start, final_end - final_start,
476 				  new_type);
477 
478 		real_updated_size += final_end - final_start;
479 
480 		/*
481 		 * left range could be head or tail, so need to update
482 		 * size at first.
483 		 */
484 		ei->size -= final_end - final_start;
485 		if (ei->addr < final_start)
486 			continue;
487 		ei->addr = final_end;
488 	}
489 	return real_updated_size;
490 }
491 
492 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
493 			     unsigned new_type)
494 {
495 	return __e820_update_range(&e820, start, size, old_type, new_type);
496 }
497 
498 static u64 __init e820_update_range_saved(u64 start, u64 size,
499 					  unsigned old_type, unsigned new_type)
500 {
501 	return __e820_update_range(&e820_saved, start, size, old_type,
502 				     new_type);
503 }
504 
505 /* make e820 not cover the range */
506 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
507 			     int checktype)
508 {
509 	int i;
510 	u64 end;
511 	u64 real_removed_size = 0;
512 
513 	if (size > (ULLONG_MAX - start))
514 		size = ULLONG_MAX - start;
515 
516 	end = start + size;
517 	printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ",
518 	       (unsigned long long) start, (unsigned long long) (end - 1));
519 	if (checktype)
520 		e820_print_type(old_type);
521 	printk(KERN_CONT "\n");
522 
523 	for (i = 0; i < e820.nr_map; i++) {
524 		struct e820entry *ei = &e820.map[i];
525 		u64 final_start, final_end;
526 		u64 ei_end;
527 
528 		if (checktype && ei->type != old_type)
529 			continue;
530 
531 		ei_end = ei->addr + ei->size;
532 		/* totally covered? */
533 		if (ei->addr >= start && ei_end <= end) {
534 			real_removed_size += ei->size;
535 			memset(ei, 0, sizeof(struct e820entry));
536 			continue;
537 		}
538 
539 		/* new range is totally covered? */
540 		if (ei->addr < start && ei_end > end) {
541 			e820_add_region(end, ei_end - end, ei->type);
542 			ei->size = start - ei->addr;
543 			real_removed_size += size;
544 			continue;
545 		}
546 
547 		/* partially covered */
548 		final_start = max(start, ei->addr);
549 		final_end = min(end, ei_end);
550 		if (final_start >= final_end)
551 			continue;
552 		real_removed_size += final_end - final_start;
553 
554 		/*
555 		 * left range could be head or tail, so need to update
556 		 * size at first.
557 		 */
558 		ei->size -= final_end - final_start;
559 		if (ei->addr < final_start)
560 			continue;
561 		ei->addr = final_end;
562 	}
563 	return real_removed_size;
564 }
565 
566 void __init update_e820(void)
567 {
568 	if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map))
569 		return;
570 	printk(KERN_INFO "e820: modified physical RAM map:\n");
571 	e820_print_map("modified");
572 }
573 static void __init update_e820_saved(void)
574 {
575 	sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map),
576 				&e820_saved.nr_map);
577 }
578 #define MAX_GAP_END 0x100000000ull
579 /*
580  * Search for a gap in the e820 memory space from start_addr to end_addr.
581  */
582 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
583 		unsigned long start_addr, unsigned long long end_addr)
584 {
585 	unsigned long long last;
586 	int i = e820.nr_map;
587 	int found = 0;
588 
589 	last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
590 
591 	while (--i >= 0) {
592 		unsigned long long start = e820.map[i].addr;
593 		unsigned long long end = start + e820.map[i].size;
594 
595 		if (end < start_addr)
596 			continue;
597 
598 		/*
599 		 * Since "last" is at most 4GB, we know we'll
600 		 * fit in 32 bits if this condition is true
601 		 */
602 		if (last > end) {
603 			unsigned long gap = last - end;
604 
605 			if (gap >= *gapsize) {
606 				*gapsize = gap;
607 				*gapstart = end;
608 				found = 1;
609 			}
610 		}
611 		if (start < last)
612 			last = start;
613 	}
614 	return found;
615 }
616 
617 /*
618  * Search for the biggest gap in the low 32 bits of the e820
619  * memory space.  We pass this space to PCI to assign MMIO resources
620  * for hotplug or unconfigured devices in.
621  * Hopefully the BIOS let enough space left.
622  */
623 __init void e820_setup_gap(void)
624 {
625 	unsigned long gapstart, gapsize;
626 	int found;
627 
628 	gapstart = 0x10000000;
629 	gapsize = 0x400000;
630 	found  = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
631 
632 #ifdef CONFIG_X86_64
633 	if (!found) {
634 		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
635 		printk(KERN_ERR
636 	"e820: cannot find a gap in the 32bit address range\n"
637 	"e820: PCI devices with unassigned 32bit BARs may break!\n");
638 	}
639 #endif
640 
641 	/*
642 	 * e820_reserve_resources_late protect stolen RAM already
643 	 */
644 	pci_mem_start = gapstart;
645 
646 	printk(KERN_INFO
647 	       "e820: [mem %#010lx-%#010lx] available for PCI devices\n",
648 	       gapstart, gapstart + gapsize - 1);
649 }
650 
651 /**
652  * Because of the size limitation of struct boot_params, only first
653  * 128 E820 memory entries are passed to kernel via
654  * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
655  * linked list of struct setup_data, which is parsed here.
656  */
657 void __init parse_e820_ext(u64 phys_addr, u32 data_len)
658 {
659 	int entries;
660 	struct e820entry *extmap;
661 	struct setup_data *sdata;
662 
663 	sdata = early_memremap(phys_addr, data_len);
664 	entries = sdata->len / sizeof(struct e820entry);
665 	extmap = (struct e820entry *)(sdata->data);
666 	__append_e820_map(extmap, entries);
667 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
668 	early_memunmap(sdata, data_len);
669 	printk(KERN_INFO "e820: extended physical RAM map:\n");
670 	e820_print_map("extended");
671 }
672 
673 #if defined(CONFIG_X86_64) || \
674 	(defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
675 /**
676  * Find the ranges of physical addresses that do not correspond to
677  * e820 RAM areas and mark the corresponding pages as nosave for
678  * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
679  *
680  * This function requires the e820 map to be sorted and without any
681  * overlapping entries.
682  */
683 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
684 {
685 	int i;
686 	unsigned long pfn = 0;
687 
688 	for (i = 0; i < e820.nr_map; i++) {
689 		struct e820entry *ei = &e820.map[i];
690 
691 		if (pfn < PFN_UP(ei->addr))
692 			register_nosave_region(pfn, PFN_UP(ei->addr));
693 
694 		pfn = PFN_DOWN(ei->addr + ei->size);
695 
696 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
697 			register_nosave_region(PFN_UP(ei->addr), pfn);
698 
699 		if (pfn >= limit_pfn)
700 			break;
701 	}
702 }
703 #endif
704 
705 #ifdef CONFIG_ACPI
706 /**
707  * Mark ACPI NVS memory region, so that we can save/restore it during
708  * hibernation and the subsequent resume.
709  */
710 static int __init e820_mark_nvs_memory(void)
711 {
712 	int i;
713 
714 	for (i = 0; i < e820.nr_map; i++) {
715 		struct e820entry *ei = &e820.map[i];
716 
717 		if (ei->type == E820_NVS)
718 			acpi_nvs_register(ei->addr, ei->size);
719 	}
720 
721 	return 0;
722 }
723 core_initcall(e820_mark_nvs_memory);
724 #endif
725 
726 /*
727  * pre allocated 4k and reserved it in memblock and e820_saved
728  */
729 u64 __init early_reserve_e820(u64 size, u64 align)
730 {
731 	u64 addr;
732 
733 	addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
734 	if (addr) {
735 		e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED);
736 		printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n");
737 		update_e820_saved();
738 	}
739 
740 	return addr;
741 }
742 
743 #ifdef CONFIG_X86_32
744 # ifdef CONFIG_X86_PAE
745 #  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
746 # else
747 #  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
748 # endif
749 #else /* CONFIG_X86_32 */
750 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
751 #endif
752 
753 /*
754  * Find the highest page frame number we have available
755  */
756 static unsigned long __init e820_end_pfn(unsigned long limit_pfn)
757 {
758 	int i;
759 	unsigned long last_pfn = 0;
760 	unsigned long max_arch_pfn = MAX_ARCH_PFN;
761 
762 	for (i = 0; i < e820.nr_map; i++) {
763 		struct e820entry *ei = &e820.map[i];
764 		unsigned long start_pfn;
765 		unsigned long end_pfn;
766 
767 		/*
768 		 * Persistent memory is accounted as ram for purposes of
769 		 * establishing max_pfn and mem_map.
770 		 */
771 		if (ei->type != E820_RAM && ei->type != E820_PRAM)
772 			continue;
773 
774 		start_pfn = ei->addr >> PAGE_SHIFT;
775 		end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
776 
777 		if (start_pfn >= limit_pfn)
778 			continue;
779 		if (end_pfn > limit_pfn) {
780 			last_pfn = limit_pfn;
781 			break;
782 		}
783 		if (end_pfn > last_pfn)
784 			last_pfn = end_pfn;
785 	}
786 
787 	if (last_pfn > max_arch_pfn)
788 		last_pfn = max_arch_pfn;
789 
790 	printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
791 			 last_pfn, max_arch_pfn);
792 	return last_pfn;
793 }
794 unsigned long __init e820_end_of_ram_pfn(void)
795 {
796 	return e820_end_pfn(MAX_ARCH_PFN);
797 }
798 
799 unsigned long __init e820_end_of_low_ram_pfn(void)
800 {
801 	return e820_end_pfn(1UL << (32-PAGE_SHIFT));
802 }
803 
804 static void early_panic(char *msg)
805 {
806 	early_printk(msg);
807 	panic(msg);
808 }
809 
810 static int userdef __initdata;
811 
812 /* "mem=nopentium" disables the 4MB page tables. */
813 static int __init parse_memopt(char *p)
814 {
815 	u64 mem_size;
816 
817 	if (!p)
818 		return -EINVAL;
819 
820 	if (!strcmp(p, "nopentium")) {
821 #ifdef CONFIG_X86_32
822 		setup_clear_cpu_cap(X86_FEATURE_PSE);
823 		return 0;
824 #else
825 		printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
826 		return -EINVAL;
827 #endif
828 	}
829 
830 	userdef = 1;
831 	mem_size = memparse(p, &p);
832 	/* don't remove all of memory when handling "mem={invalid}" param */
833 	if (mem_size == 0)
834 		return -EINVAL;
835 	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
836 
837 	return 0;
838 }
839 early_param("mem", parse_memopt);
840 
841 static int __init parse_memmap_one(char *p)
842 {
843 	char *oldp;
844 	u64 start_at, mem_size;
845 
846 	if (!p)
847 		return -EINVAL;
848 
849 	if (!strncmp(p, "exactmap", 8)) {
850 #ifdef CONFIG_CRASH_DUMP
851 		/*
852 		 * If we are doing a crash dump, we still need to know
853 		 * the real mem size before original memory map is
854 		 * reset.
855 		 */
856 		saved_max_pfn = e820_end_of_ram_pfn();
857 #endif
858 		e820.nr_map = 0;
859 		userdef = 1;
860 		return 0;
861 	}
862 
863 	oldp = p;
864 	mem_size = memparse(p, &p);
865 	if (p == oldp)
866 		return -EINVAL;
867 
868 	userdef = 1;
869 	if (*p == '@') {
870 		start_at = memparse(p+1, &p);
871 		e820_add_region(start_at, mem_size, E820_RAM);
872 	} else if (*p == '#') {
873 		start_at = memparse(p+1, &p);
874 		e820_add_region(start_at, mem_size, E820_ACPI);
875 	} else if (*p == '$') {
876 		start_at = memparse(p+1, &p);
877 		e820_add_region(start_at, mem_size, E820_RESERVED);
878 	} else if (*p == '!') {
879 		start_at = memparse(p+1, &p);
880 		e820_add_region(start_at, mem_size, E820_PRAM);
881 	} else
882 		e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
883 
884 	return *p == '\0' ? 0 : -EINVAL;
885 }
886 static int __init parse_memmap_opt(char *str)
887 {
888 	while (str) {
889 		char *k = strchr(str, ',');
890 
891 		if (k)
892 			*k++ = 0;
893 
894 		parse_memmap_one(str);
895 		str = k;
896 	}
897 
898 	return 0;
899 }
900 early_param("memmap", parse_memmap_opt);
901 
902 void __init finish_e820_parsing(void)
903 {
904 	if (userdef) {
905 		if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map),
906 					&e820.nr_map) < 0)
907 			early_panic("Invalid user supplied memory map");
908 
909 		printk(KERN_INFO "e820: user-defined physical RAM map:\n");
910 		e820_print_map("user");
911 	}
912 }
913 
914 static inline const char *e820_type_to_string(int e820_type)
915 {
916 	switch (e820_type) {
917 	case E820_RESERVED_KERN:
918 	case E820_RAM:	return "System RAM";
919 	case E820_ACPI:	return "ACPI Tables";
920 	case E820_NVS:	return "ACPI Non-volatile Storage";
921 	case E820_UNUSABLE:	return "Unusable memory";
922 	case E820_PRAM: return "Persistent Memory (legacy)";
923 	case E820_PMEM: return "Persistent Memory";
924 	default:	return "reserved";
925 	}
926 }
927 
928 static bool do_mark_busy(u32 type, struct resource *res)
929 {
930 	/* this is the legacy bios/dos rom-shadow + mmio region */
931 	if (res->start < (1ULL<<20))
932 		return true;
933 
934 	/*
935 	 * Treat persistent memory like device memory, i.e. reserve it
936 	 * for exclusive use of a driver
937 	 */
938 	switch (type) {
939 	case E820_RESERVED:
940 	case E820_PRAM:
941 	case E820_PMEM:
942 		return false;
943 	default:
944 		return true;
945 	}
946 }
947 
948 /*
949  * Mark e820 reserved areas as busy for the resource manager.
950  */
951 static struct resource __initdata *e820_res;
952 void __init e820_reserve_resources(void)
953 {
954 	int i;
955 	struct resource *res;
956 	u64 end;
957 
958 	res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
959 	e820_res = res;
960 	for (i = 0; i < e820.nr_map; i++) {
961 		end = e820.map[i].addr + e820.map[i].size - 1;
962 		if (end != (resource_size_t)end) {
963 			res++;
964 			continue;
965 		}
966 		res->name = e820_type_to_string(e820.map[i].type);
967 		res->start = e820.map[i].addr;
968 		res->end = end;
969 
970 		res->flags = IORESOURCE_MEM;
971 
972 		/*
973 		 * don't register the region that could be conflicted with
974 		 * pci device BAR resource and insert them later in
975 		 * pcibios_resource_survey()
976 		 */
977 		if (do_mark_busy(e820.map[i].type, res)) {
978 			res->flags |= IORESOURCE_BUSY;
979 			insert_resource(&iomem_resource, res);
980 		}
981 		res++;
982 	}
983 
984 	for (i = 0; i < e820_saved.nr_map; i++) {
985 		struct e820entry *entry = &e820_saved.map[i];
986 		firmware_map_add_early(entry->addr,
987 			entry->addr + entry->size,
988 			e820_type_to_string(entry->type));
989 	}
990 }
991 
992 /* How much should we pad RAM ending depending on where it is? */
993 static unsigned long ram_alignment(resource_size_t pos)
994 {
995 	unsigned long mb = pos >> 20;
996 
997 	/* To 64kB in the first megabyte */
998 	if (!mb)
999 		return 64*1024;
1000 
1001 	/* To 1MB in the first 16MB */
1002 	if (mb < 16)
1003 		return 1024*1024;
1004 
1005 	/* To 64MB for anything above that */
1006 	return 64*1024*1024;
1007 }
1008 
1009 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1010 
1011 void __init e820_reserve_resources_late(void)
1012 {
1013 	int i;
1014 	struct resource *res;
1015 
1016 	res = e820_res;
1017 	for (i = 0; i < e820.nr_map; i++) {
1018 		if (!res->parent && res->end)
1019 			insert_resource_expand_to_fit(&iomem_resource, res);
1020 		res++;
1021 	}
1022 
1023 	/*
1024 	 * Try to bump up RAM regions to reasonable boundaries to
1025 	 * avoid stolen RAM:
1026 	 */
1027 	for (i = 0; i < e820.nr_map; i++) {
1028 		struct e820entry *entry = &e820.map[i];
1029 		u64 start, end;
1030 
1031 		if (entry->type != E820_RAM)
1032 			continue;
1033 		start = entry->addr + entry->size;
1034 		end = round_up(start, ram_alignment(start)) - 1;
1035 		if (end > MAX_RESOURCE_SIZE)
1036 			end = MAX_RESOURCE_SIZE;
1037 		if (start >= end)
1038 			continue;
1039 		printk(KERN_DEBUG
1040 		       "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n",
1041 		       start, end);
1042 		reserve_region_with_split(&iomem_resource, start, end,
1043 					  "RAM buffer");
1044 	}
1045 }
1046 
1047 char *__init default_machine_specific_memory_setup(void)
1048 {
1049 	char *who = "BIOS-e820";
1050 	u32 new_nr;
1051 	/*
1052 	 * Try to copy the BIOS-supplied E820-map.
1053 	 *
1054 	 * Otherwise fake a memory map; one section from 0k->640k,
1055 	 * the next section from 1mb->appropriate_mem_k
1056 	 */
1057 	new_nr = boot_params.e820_entries;
1058 	sanitize_e820_map(boot_params.e820_map,
1059 			ARRAY_SIZE(boot_params.e820_map),
1060 			&new_nr);
1061 	boot_params.e820_entries = new_nr;
1062 	if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1063 	  < 0) {
1064 		u64 mem_size;
1065 
1066 		/* compare results from other methods and take the greater */
1067 		if (boot_params.alt_mem_k
1068 		    < boot_params.screen_info.ext_mem_k) {
1069 			mem_size = boot_params.screen_info.ext_mem_k;
1070 			who = "BIOS-88";
1071 		} else {
1072 			mem_size = boot_params.alt_mem_k;
1073 			who = "BIOS-e801";
1074 		}
1075 
1076 		e820.nr_map = 0;
1077 		e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1078 		e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1079 	}
1080 
1081 	/* In case someone cares... */
1082 	return who;
1083 }
1084 
1085 void __init setup_memory_map(void)
1086 {
1087 	char *who;
1088 
1089 	who = x86_init.resources.memory_setup();
1090 	memcpy(&e820_saved, &e820, sizeof(struct e820map));
1091 	printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n");
1092 	e820_print_map(who);
1093 }
1094 
1095 void __init memblock_x86_fill(void)
1096 {
1097 	int i;
1098 	u64 end;
1099 
1100 	/*
1101 	 * EFI may have more than 128 entries
1102 	 * We are safe to enable resizing, beause memblock_x86_fill()
1103 	 * is rather later for x86
1104 	 */
1105 	memblock_allow_resize();
1106 
1107 	for (i = 0; i < e820.nr_map; i++) {
1108 		struct e820entry *ei = &e820.map[i];
1109 
1110 		end = ei->addr + ei->size;
1111 		if (end != (resource_size_t)end)
1112 			continue;
1113 
1114 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1115 			continue;
1116 
1117 		memblock_add(ei->addr, ei->size);
1118 	}
1119 
1120 	/* throw away partial pages */
1121 	memblock_trim_memory(PAGE_SIZE);
1122 
1123 	memblock_dump_all();
1124 }
1125 
1126 void __init memblock_find_dma_reserve(void)
1127 {
1128 #ifdef CONFIG_X86_64
1129 	u64 nr_pages = 0, nr_free_pages = 0;
1130 	unsigned long start_pfn, end_pfn;
1131 	phys_addr_t start, end;
1132 	int i;
1133 	u64 u;
1134 
1135 	/*
1136 	 * need to find out used area below MAX_DMA_PFN
1137 	 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1138 	 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1139 	 */
1140 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1141 		start_pfn = min(start_pfn, MAX_DMA_PFN);
1142 		end_pfn = min(end_pfn, MAX_DMA_PFN);
1143 		nr_pages += end_pfn - start_pfn;
1144 	}
1145 
1146 	for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1147 				NULL) {
1148 		start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1149 		end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1150 		if (start_pfn < end_pfn)
1151 			nr_free_pages += end_pfn - start_pfn;
1152 	}
1153 
1154 	set_dma_reserve(nr_pages - nr_free_pages);
1155 #endif
1156 }
1157