xref: /linux/arch/x86/kernel/e820.c (revision 95db3b255fde4e830e5f8cc011eb404023f669d4)
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  */
43 struct e820map e820;
44 struct e820map e820_saved;
45 
46 /* For PCI or other memory-mapped resources */
47 unsigned long pci_mem_start = 0xaeedbabe;
48 #ifdef CONFIG_PCI
49 EXPORT_SYMBOL(pci_mem_start);
50 #endif
51 
52 /*
53  * This function checks if any part of the range <start,end> is mapped
54  * with type.
55  */
56 int
57 e820_any_mapped(u64 start, u64 end, unsigned type)
58 {
59 	int i;
60 
61 	for (i = 0; i < e820.nr_map; i++) {
62 		struct e820entry *ei = &e820.map[i];
63 
64 		if (type && ei->type != type)
65 			continue;
66 		if (ei->addr >= end || ei->addr + ei->size <= start)
67 			continue;
68 		return 1;
69 	}
70 	return 0;
71 }
72 EXPORT_SYMBOL_GPL(e820_any_mapped);
73 
74 /*
75  * This function checks if the entire range <start,end> is mapped with type.
76  *
77  * Note: this function only works correct if the e820 table is sorted and
78  * not-overlapping, which is the case
79  */
80 int __init e820_all_mapped(u64 start, u64 end, unsigned type)
81 {
82 	int i;
83 
84 	for (i = 0; i < e820.nr_map; i++) {
85 		struct e820entry *ei = &e820.map[i];
86 
87 		if (type && ei->type != type)
88 			continue;
89 		/* is the region (part) in overlap with the current region ?*/
90 		if (ei->addr >= end || ei->addr + ei->size <= start)
91 			continue;
92 
93 		/* if the region is at the beginning of <start,end> we move
94 		 * start to the end of the region since it's ok until there
95 		 */
96 		if (ei->addr <= start)
97 			start = ei->addr + ei->size;
98 		/*
99 		 * if start is now at or beyond end, we're done, full
100 		 * coverage
101 		 */
102 		if (start >= end)
103 			return 1;
104 	}
105 	return 0;
106 }
107 
108 /*
109  * Add a memory region to the kernel e820 map.
110  */
111 static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
112 					 int type)
113 {
114 	int x = e820x->nr_map;
115 
116 	if (x >= ARRAY_SIZE(e820x->map)) {
117 		printk(KERN_ERR "e820: too many entries; ignoring [mem %#010llx-%#010llx]\n",
118 		       (unsigned long long) start,
119 		       (unsigned long long) (start + size - 1));
120 		return;
121 	}
122 
123 	e820x->map[x].addr = start;
124 	e820x->map[x].size = size;
125 	e820x->map[x].type = type;
126 	e820x->nr_map++;
127 }
128 
129 void __init e820_add_region(u64 start, u64 size, int type)
130 {
131 	__e820_add_region(&e820, start, size, type);
132 }
133 
134 static void __init e820_print_type(u32 type)
135 {
136 	switch (type) {
137 	case E820_RAM:
138 	case E820_RESERVED_KERN:
139 		printk(KERN_CONT "usable");
140 		break;
141 	case E820_RESERVED:
142 		printk(KERN_CONT "reserved");
143 		break;
144 	case E820_ACPI:
145 		printk(KERN_CONT "ACPI data");
146 		break;
147 	case E820_NVS:
148 		printk(KERN_CONT "ACPI NVS");
149 		break;
150 	case E820_UNUSABLE:
151 		printk(KERN_CONT "unusable");
152 		break;
153 	case E820_PMEM:
154 	case E820_PRAM:
155 		printk(KERN_CONT "persistent (type %u)", type);
156 		break;
157 	default:
158 		printk(KERN_CONT "type %u", type);
159 		break;
160 	}
161 }
162 
163 void __init e820_print_map(char *who)
164 {
165 	int i;
166 
167 	for (i = 0; i < e820.nr_map; i++) {
168 		printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who,
169 		       (unsigned long long) e820.map[i].addr,
170 		       (unsigned long long)
171 		       (e820.map[i].addr + e820.map[i].size - 1));
172 		e820_print_type(e820.map[i].type);
173 		printk(KERN_CONT "\n");
174 	}
175 }
176 
177 /*
178  * Sanitize the BIOS e820 map.
179  *
180  * Some e820 responses include overlapping entries. The following
181  * replaces the original e820 map with a new one, removing overlaps,
182  * and resolving conflicting memory types in favor of highest
183  * numbered type.
184  *
185  * The input parameter biosmap points to an array of 'struct
186  * e820entry' which on entry has elements in the range [0, *pnr_map)
187  * valid, and which has space for up to max_nr_map entries.
188  * On return, the resulting sanitized e820 map entries will be in
189  * overwritten in the same location, starting at biosmap.
190  *
191  * The integer pointed to by pnr_map must be valid on entry (the
192  * current number of valid entries located at biosmap). If the
193  * sanitizing succeeds the *pnr_map will be updated with the new
194  * number of valid entries (something no more than max_nr_map).
195  *
196  * The return value from sanitize_e820_map() is zero if it
197  * successfully 'sanitized' the map entries passed in, and is -1
198  * if it did nothing, which can happen if either of (1) it was
199  * only passed one map entry, or (2) any of the input map entries
200  * were invalid (start + size < start, meaning that the size was
201  * so big the described memory range wrapped around through zero.)
202  *
203  *	Visually we're performing the following
204  *	(1,2,3,4 = memory types)...
205  *
206  *	Sample memory map (w/overlaps):
207  *	   ____22__________________
208  *	   ______________________4_
209  *	   ____1111________________
210  *	   _44_____________________
211  *	   11111111________________
212  *	   ____________________33__
213  *	   ___________44___________
214  *	   __________33333_________
215  *	   ______________22________
216  *	   ___________________2222_
217  *	   _________111111111______
218  *	   _____________________11_
219  *	   _________________4______
220  *
221  *	Sanitized equivalent (no overlap):
222  *	   1_______________________
223  *	   _44_____________________
224  *	   ___1____________________
225  *	   ____22__________________
226  *	   ______11________________
227  *	   _________1______________
228  *	   __________3_____________
229  *	   ___________44___________
230  *	   _____________33_________
231  *	   _______________2________
232  *	   ________________1_______
233  *	   _________________4______
234  *	   ___________________2____
235  *	   ____________________33__
236  *	   ______________________4_
237  */
238 struct change_member {
239 	struct e820entry *pbios; /* pointer to original bios entry */
240 	unsigned long long addr; /* address for this change point */
241 };
242 
243 static int __init cpcompare(const void *a, const void *b)
244 {
245 	struct change_member * const *app = a, * const *bpp = b;
246 	const struct change_member *ap = *app, *bp = *bpp;
247 
248 	/*
249 	 * Inputs are pointers to two elements of change_point[].  If their
250 	 * addresses are unequal, their difference dominates.  If the addresses
251 	 * are equal, then consider one that represents the end of its region
252 	 * to be greater than one that does not.
253 	 */
254 	if (ap->addr != bp->addr)
255 		return ap->addr > bp->addr ? 1 : -1;
256 
257 	return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr);
258 }
259 
260 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
261 			     u32 *pnr_map)
262 {
263 	static struct change_member change_point_list[2*E820_X_MAX] __initdata;
264 	static struct change_member *change_point[2*E820_X_MAX] __initdata;
265 	static struct e820entry *overlap_list[E820_X_MAX] __initdata;
266 	static struct e820entry new_bios[E820_X_MAX] __initdata;
267 	unsigned long current_type, last_type;
268 	unsigned long long last_addr;
269 	int chgidx;
270 	int overlap_entries;
271 	int new_bios_entry;
272 	int old_nr, new_nr, chg_nr;
273 	int i;
274 
275 	/* if there's only one memory region, don't bother */
276 	if (*pnr_map < 2)
277 		return -1;
278 
279 	old_nr = *pnr_map;
280 	BUG_ON(old_nr > max_nr_map);
281 
282 	/* bail out if we find any unreasonable addresses in bios map */
283 	for (i = 0; i < old_nr; i++)
284 		if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
285 			return -1;
286 
287 	/* create pointers for initial change-point information (for sorting) */
288 	for (i = 0; i < 2 * old_nr; i++)
289 		change_point[i] = &change_point_list[i];
290 
291 	/* record all known change-points (starting and ending addresses),
292 	   omitting those that are for empty memory regions */
293 	chgidx = 0;
294 	for (i = 0; i < old_nr; i++)	{
295 		if (biosmap[i].size != 0) {
296 			change_point[chgidx]->addr = biosmap[i].addr;
297 			change_point[chgidx++]->pbios = &biosmap[i];
298 			change_point[chgidx]->addr = biosmap[i].addr +
299 				biosmap[i].size;
300 			change_point[chgidx++]->pbios = &biosmap[i];
301 		}
302 	}
303 	chg_nr = chgidx;
304 
305 	/* sort change-point list by memory addresses (low -> high) */
306 	sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL);
307 
308 	/* create a new bios memory map, removing overlaps */
309 	overlap_entries = 0;	 /* number of entries in the overlap table */
310 	new_bios_entry = 0;	 /* index for creating new bios map entries */
311 	last_type = 0;		 /* start with undefined memory type */
312 	last_addr = 0;		 /* start with 0 as last starting address */
313 
314 	/* loop through change-points, determining affect on the new bios map */
315 	for (chgidx = 0; chgidx < chg_nr; chgidx++) {
316 		/* keep track of all overlapping bios entries */
317 		if (change_point[chgidx]->addr ==
318 		    change_point[chgidx]->pbios->addr) {
319 			/*
320 			 * add map entry to overlap list (> 1 entry
321 			 * implies an overlap)
322 			 */
323 			overlap_list[overlap_entries++] =
324 				change_point[chgidx]->pbios;
325 		} else {
326 			/*
327 			 * remove entry from list (order independent,
328 			 * so swap with last)
329 			 */
330 			for (i = 0; i < overlap_entries; i++) {
331 				if (overlap_list[i] ==
332 				    change_point[chgidx]->pbios)
333 					overlap_list[i] =
334 						overlap_list[overlap_entries-1];
335 			}
336 			overlap_entries--;
337 		}
338 		/*
339 		 * if there are overlapping entries, decide which
340 		 * "type" to use (larger value takes precedence --
341 		 * 1=usable, 2,3,4,4+=unusable)
342 		 */
343 		current_type = 0;
344 		for (i = 0; i < overlap_entries; i++)
345 			if (overlap_list[i]->type > current_type)
346 				current_type = overlap_list[i]->type;
347 		/*
348 		 * continue building up new bios map based on this
349 		 * information
350 		 */
351 		if (current_type != last_type || current_type == E820_PRAM) {
352 			if (last_type != 0)	 {
353 				new_bios[new_bios_entry].size =
354 					change_point[chgidx]->addr - last_addr;
355 				/*
356 				 * move forward only if the new size
357 				 * was non-zero
358 				 */
359 				if (new_bios[new_bios_entry].size != 0)
360 					/*
361 					 * no more space left for new
362 					 * bios entries ?
363 					 */
364 					if (++new_bios_entry >= max_nr_map)
365 						break;
366 			}
367 			if (current_type != 0)	{
368 				new_bios[new_bios_entry].addr =
369 					change_point[chgidx]->addr;
370 				new_bios[new_bios_entry].type = current_type;
371 				last_addr = change_point[chgidx]->addr;
372 			}
373 			last_type = current_type;
374 		}
375 	}
376 	/* retain count for new bios entries */
377 	new_nr = new_bios_entry;
378 
379 	/* copy new bios mapping into original location */
380 	memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
381 	*pnr_map = new_nr;
382 
383 	return 0;
384 }
385 
386 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
387 {
388 	while (nr_map) {
389 		u64 start = biosmap->addr;
390 		u64 size = biosmap->size;
391 		u64 end = start + size;
392 		u32 type = biosmap->type;
393 
394 		/* Overflow in 64 bits? Ignore the memory map. */
395 		if (start > end)
396 			return -1;
397 
398 		e820_add_region(start, size, type);
399 
400 		biosmap++;
401 		nr_map--;
402 	}
403 	return 0;
404 }
405 
406 /*
407  * Copy the BIOS e820 map into a safe place.
408  *
409  * Sanity-check it while we're at it..
410  *
411  * If we're lucky and live on a modern system, the setup code
412  * will have given us a memory map that we can use to properly
413  * set up memory.  If we aren't, we'll fake a memory map.
414  */
415 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
416 {
417 	/* Only one memory region (or negative)? Ignore it */
418 	if (nr_map < 2)
419 		return -1;
420 
421 	return __append_e820_map(biosmap, nr_map);
422 }
423 
424 static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
425 					u64 size, unsigned old_type,
426 					unsigned new_type)
427 {
428 	u64 end;
429 	unsigned int i;
430 	u64 real_updated_size = 0;
431 
432 	BUG_ON(old_type == new_type);
433 
434 	if (size > (ULLONG_MAX - start))
435 		size = ULLONG_MAX - start;
436 
437 	end = start + size;
438 	printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ",
439 	       (unsigned long long) start, (unsigned long long) (end - 1));
440 	e820_print_type(old_type);
441 	printk(KERN_CONT " ==> ");
442 	e820_print_type(new_type);
443 	printk(KERN_CONT "\n");
444 
445 	for (i = 0; i < e820x->nr_map; i++) {
446 		struct e820entry *ei = &e820x->map[i];
447 		u64 final_start, final_end;
448 		u64 ei_end;
449 
450 		if (ei->type != old_type)
451 			continue;
452 
453 		ei_end = ei->addr + ei->size;
454 		/* totally covered by new range? */
455 		if (ei->addr >= start && ei_end <= end) {
456 			ei->type = new_type;
457 			real_updated_size += ei->size;
458 			continue;
459 		}
460 
461 		/* new range is totally covered? */
462 		if (ei->addr < start && ei_end > end) {
463 			__e820_add_region(e820x, start, size, new_type);
464 			__e820_add_region(e820x, end, ei_end - end, ei->type);
465 			ei->size = start - ei->addr;
466 			real_updated_size += size;
467 			continue;
468 		}
469 
470 		/* partially covered */
471 		final_start = max(start, ei->addr);
472 		final_end = min(end, ei_end);
473 		if (final_start >= final_end)
474 			continue;
475 
476 		__e820_add_region(e820x, final_start, final_end - final_start,
477 				  new_type);
478 
479 		real_updated_size += final_end - final_start;
480 
481 		/*
482 		 * left range could be head or tail, so need to update
483 		 * size at first.
484 		 */
485 		ei->size -= final_end - final_start;
486 		if (ei->addr < final_start)
487 			continue;
488 		ei->addr = final_end;
489 	}
490 	return real_updated_size;
491 }
492 
493 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
494 			     unsigned new_type)
495 {
496 	return __e820_update_range(&e820, start, size, old_type, new_type);
497 }
498 
499 static u64 __init e820_update_range_saved(u64 start, u64 size,
500 					  unsigned old_type, unsigned new_type)
501 {
502 	return __e820_update_range(&e820_saved, start, size, old_type,
503 				     new_type);
504 }
505 
506 /* make e820 not cover the range */
507 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
508 			     int checktype)
509 {
510 	int i;
511 	u64 end;
512 	u64 real_removed_size = 0;
513 
514 	if (size > (ULLONG_MAX - start))
515 		size = ULLONG_MAX - start;
516 
517 	end = start + size;
518 	printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ",
519 	       (unsigned long long) start, (unsigned long long) (end - 1));
520 	if (checktype)
521 		e820_print_type(old_type);
522 	printk(KERN_CONT "\n");
523 
524 	for (i = 0; i < e820.nr_map; i++) {
525 		struct e820entry *ei = &e820.map[i];
526 		u64 final_start, final_end;
527 		u64 ei_end;
528 
529 		if (checktype && ei->type != old_type)
530 			continue;
531 
532 		ei_end = ei->addr + ei->size;
533 		/* totally covered? */
534 		if (ei->addr >= start && ei_end <= end) {
535 			real_removed_size += ei->size;
536 			memset(ei, 0, sizeof(struct e820entry));
537 			continue;
538 		}
539 
540 		/* new range is totally covered? */
541 		if (ei->addr < start && ei_end > end) {
542 			e820_add_region(end, ei_end - end, ei->type);
543 			ei->size = start - ei->addr;
544 			real_removed_size += size;
545 			continue;
546 		}
547 
548 		/* partially covered */
549 		final_start = max(start, ei->addr);
550 		final_end = min(end, ei_end);
551 		if (final_start >= final_end)
552 			continue;
553 		real_removed_size += final_end - final_start;
554 
555 		/*
556 		 * left range could be head or tail, so need to update
557 		 * size at first.
558 		 */
559 		ei->size -= final_end - final_start;
560 		if (ei->addr < final_start)
561 			continue;
562 		ei->addr = final_end;
563 	}
564 	return real_removed_size;
565 }
566 
567 void __init update_e820(void)
568 {
569 	if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map))
570 		return;
571 	printk(KERN_INFO "e820: modified physical RAM map:\n");
572 	e820_print_map("modified");
573 }
574 static void __init update_e820_saved(void)
575 {
576 	sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map),
577 				&e820_saved.nr_map);
578 }
579 #define MAX_GAP_END 0x100000000ull
580 /*
581  * Search for a gap in the e820 memory space from start_addr to end_addr.
582  */
583 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
584 		unsigned long start_addr, unsigned long long end_addr)
585 {
586 	unsigned long long last;
587 	int i = e820.nr_map;
588 	int found = 0;
589 
590 	last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
591 
592 	while (--i >= 0) {
593 		unsigned long long start = e820.map[i].addr;
594 		unsigned long long end = start + e820.map[i].size;
595 
596 		if (end < start_addr)
597 			continue;
598 
599 		/*
600 		 * Since "last" is at most 4GB, we know we'll
601 		 * fit in 32 bits if this condition is true
602 		 */
603 		if (last > end) {
604 			unsigned long gap = last - end;
605 
606 			if (gap >= *gapsize) {
607 				*gapsize = gap;
608 				*gapstart = end;
609 				found = 1;
610 			}
611 		}
612 		if (start < last)
613 			last = start;
614 	}
615 	return found;
616 }
617 
618 /*
619  * Search for the biggest gap in the low 32 bits of the e820
620  * memory space.  We pass this space to PCI to assign MMIO resources
621  * for hotplug or unconfigured devices in.
622  * Hopefully the BIOS let enough space left.
623  */
624 __init void e820_setup_gap(void)
625 {
626 	unsigned long gapstart, gapsize;
627 	int found;
628 
629 	gapstart = 0x10000000;
630 	gapsize = 0x400000;
631 	found  = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
632 
633 #ifdef CONFIG_X86_64
634 	if (!found) {
635 		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
636 		printk(KERN_ERR
637 	"e820: cannot find a gap in the 32bit address range\n"
638 	"e820: PCI devices with unassigned 32bit BARs may break!\n");
639 	}
640 #endif
641 
642 	/*
643 	 * e820_reserve_resources_late protect stolen RAM already
644 	 */
645 	pci_mem_start = gapstart;
646 
647 	printk(KERN_INFO
648 	       "e820: [mem %#010lx-%#010lx] available for PCI devices\n",
649 	       gapstart, gapstart + gapsize - 1);
650 }
651 
652 /**
653  * Because of the size limitation of struct boot_params, only first
654  * 128 E820 memory entries are passed to kernel via
655  * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
656  * linked list of struct setup_data, which is parsed here.
657  */
658 void __init parse_e820_ext(u64 phys_addr, u32 data_len)
659 {
660 	int entries;
661 	struct e820entry *extmap;
662 	struct setup_data *sdata;
663 
664 	sdata = early_memremap(phys_addr, data_len);
665 	entries = sdata->len / sizeof(struct e820entry);
666 	extmap = (struct e820entry *)(sdata->data);
667 	__append_e820_map(extmap, entries);
668 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
669 	early_memunmap(sdata, data_len);
670 	printk(KERN_INFO "e820: extended physical RAM map:\n");
671 	e820_print_map("extended");
672 }
673 
674 #if defined(CONFIG_X86_64) || \
675 	(defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
676 /**
677  * Find the ranges of physical addresses that do not correspond to
678  * e820 RAM areas and mark the corresponding pages as nosave for
679  * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
680  *
681  * This function requires the e820 map to be sorted and without any
682  * overlapping entries.
683  */
684 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
685 {
686 	int i;
687 	unsigned long pfn = 0;
688 
689 	for (i = 0; i < e820.nr_map; i++) {
690 		struct e820entry *ei = &e820.map[i];
691 
692 		if (pfn < PFN_UP(ei->addr))
693 			register_nosave_region(pfn, PFN_UP(ei->addr));
694 
695 		pfn = PFN_DOWN(ei->addr + ei->size);
696 
697 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
698 			register_nosave_region(PFN_UP(ei->addr), pfn);
699 
700 		if (pfn >= limit_pfn)
701 			break;
702 	}
703 }
704 #endif
705 
706 #ifdef CONFIG_ACPI
707 /**
708  * Mark ACPI NVS memory region, so that we can save/restore it during
709  * hibernation and the subsequent resume.
710  */
711 static int __init e820_mark_nvs_memory(void)
712 {
713 	int i;
714 
715 	for (i = 0; i < e820.nr_map; i++) {
716 		struct e820entry *ei = &e820.map[i];
717 
718 		if (ei->type == E820_NVS)
719 			acpi_nvs_register(ei->addr, ei->size);
720 	}
721 
722 	return 0;
723 }
724 core_initcall(e820_mark_nvs_memory);
725 #endif
726 
727 /*
728  * pre allocated 4k and reserved it in memblock and e820_saved
729  */
730 u64 __init early_reserve_e820(u64 size, u64 align)
731 {
732 	u64 addr;
733 
734 	addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
735 	if (addr) {
736 		e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED);
737 		printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n");
738 		update_e820_saved();
739 	}
740 
741 	return addr;
742 }
743 
744 #ifdef CONFIG_X86_32
745 # ifdef CONFIG_X86_PAE
746 #  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
747 # else
748 #  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
749 # endif
750 #else /* CONFIG_X86_32 */
751 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
752 #endif
753 
754 /*
755  * Find the highest page frame number we have available
756  */
757 static unsigned long __init e820_end_pfn(unsigned long limit_pfn)
758 {
759 	int i;
760 	unsigned long last_pfn = 0;
761 	unsigned long max_arch_pfn = MAX_ARCH_PFN;
762 
763 	for (i = 0; i < e820.nr_map; i++) {
764 		struct e820entry *ei = &e820.map[i];
765 		unsigned long start_pfn;
766 		unsigned long end_pfn;
767 
768 		/*
769 		 * Persistent memory is accounted as ram for purposes of
770 		 * establishing max_pfn and mem_map.
771 		 */
772 		if (ei->type != E820_RAM && ei->type != E820_PRAM)
773 			continue;
774 
775 		start_pfn = ei->addr >> PAGE_SHIFT;
776 		end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
777 
778 		if (start_pfn >= limit_pfn)
779 			continue;
780 		if (end_pfn > limit_pfn) {
781 			last_pfn = limit_pfn;
782 			break;
783 		}
784 		if (end_pfn > last_pfn)
785 			last_pfn = end_pfn;
786 	}
787 
788 	if (last_pfn > max_arch_pfn)
789 		last_pfn = max_arch_pfn;
790 
791 	printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
792 			 last_pfn, max_arch_pfn);
793 	return last_pfn;
794 }
795 unsigned long __init e820_end_of_ram_pfn(void)
796 {
797 	return e820_end_pfn(MAX_ARCH_PFN);
798 }
799 
800 unsigned long __init e820_end_of_low_ram_pfn(void)
801 {
802 	return e820_end_pfn(1UL << (32-PAGE_SHIFT));
803 }
804 
805 static void early_panic(char *msg)
806 {
807 	early_printk(msg);
808 	panic(msg);
809 }
810 
811 static int userdef __initdata;
812 
813 /* "mem=nopentium" disables the 4MB page tables. */
814 static int __init parse_memopt(char *p)
815 {
816 	u64 mem_size;
817 
818 	if (!p)
819 		return -EINVAL;
820 
821 	if (!strcmp(p, "nopentium")) {
822 #ifdef CONFIG_X86_32
823 		setup_clear_cpu_cap(X86_FEATURE_PSE);
824 		return 0;
825 #else
826 		printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
827 		return -EINVAL;
828 #endif
829 	}
830 
831 	userdef = 1;
832 	mem_size = memparse(p, &p);
833 	/* don't remove all of memory when handling "mem={invalid}" param */
834 	if (mem_size == 0)
835 		return -EINVAL;
836 	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
837 
838 	return 0;
839 }
840 early_param("mem", parse_memopt);
841 
842 static int __init parse_memmap_one(char *p)
843 {
844 	char *oldp;
845 	u64 start_at, mem_size;
846 
847 	if (!p)
848 		return -EINVAL;
849 
850 	if (!strncmp(p, "exactmap", 8)) {
851 #ifdef CONFIG_CRASH_DUMP
852 		/*
853 		 * If we are doing a crash dump, we still need to know
854 		 * the real mem size before original memory map is
855 		 * reset.
856 		 */
857 		saved_max_pfn = e820_end_of_ram_pfn();
858 #endif
859 		e820.nr_map = 0;
860 		userdef = 1;
861 		return 0;
862 	}
863 
864 	oldp = p;
865 	mem_size = memparse(p, &p);
866 	if (p == oldp)
867 		return -EINVAL;
868 
869 	userdef = 1;
870 	if (*p == '@') {
871 		start_at = memparse(p+1, &p);
872 		e820_add_region(start_at, mem_size, E820_RAM);
873 	} else if (*p == '#') {
874 		start_at = memparse(p+1, &p);
875 		e820_add_region(start_at, mem_size, E820_ACPI);
876 	} else if (*p == '$') {
877 		start_at = memparse(p+1, &p);
878 		e820_add_region(start_at, mem_size, E820_RESERVED);
879 	} else if (*p == '!') {
880 		start_at = memparse(p+1, &p);
881 		e820_add_region(start_at, mem_size, E820_PRAM);
882 	} else
883 		e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
884 
885 	return *p == '\0' ? 0 : -EINVAL;
886 }
887 static int __init parse_memmap_opt(char *str)
888 {
889 	while (str) {
890 		char *k = strchr(str, ',');
891 
892 		if (k)
893 			*k++ = 0;
894 
895 		parse_memmap_one(str);
896 		str = k;
897 	}
898 
899 	return 0;
900 }
901 early_param("memmap", parse_memmap_opt);
902 
903 void __init finish_e820_parsing(void)
904 {
905 	if (userdef) {
906 		if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map),
907 					&e820.nr_map) < 0)
908 			early_panic("Invalid user supplied memory map");
909 
910 		printk(KERN_INFO "e820: user-defined physical RAM map:\n");
911 		e820_print_map("user");
912 	}
913 }
914 
915 static const char *e820_type_to_string(int e820_type)
916 {
917 	switch (e820_type) {
918 	case E820_RESERVED_KERN:
919 	case E820_RAM:	return "System RAM";
920 	case E820_ACPI:	return "ACPI Tables";
921 	case E820_NVS:	return "ACPI Non-volatile Storage";
922 	case E820_UNUSABLE:	return "Unusable memory";
923 	case E820_PRAM: return "Persistent Memory (legacy)";
924 	case E820_PMEM: return "Persistent Memory";
925 	default:	return "reserved";
926 	}
927 }
928 
929 static unsigned long e820_type_to_iomem_type(int e820_type)
930 {
931 	switch (e820_type) {
932 	case E820_RESERVED_KERN:
933 	case E820_RAM:
934 		return IORESOURCE_SYSTEM_RAM;
935 	case E820_ACPI:
936 	case E820_NVS:
937 	case E820_UNUSABLE:
938 	case E820_PRAM:
939 	case E820_PMEM:
940 	default:
941 		return IORESOURCE_MEM;
942 	}
943 }
944 
945 static unsigned long e820_type_to_iores_desc(int e820_type)
946 {
947 	switch (e820_type) {
948 	case E820_ACPI:
949 		return IORES_DESC_ACPI_TABLES;
950 	case E820_NVS:
951 		return IORES_DESC_ACPI_NV_STORAGE;
952 	case E820_PMEM:
953 		return IORES_DESC_PERSISTENT_MEMORY;
954 	case E820_PRAM:
955 		return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
956 	case E820_RESERVED_KERN:
957 	case E820_RAM:
958 	case E820_UNUSABLE:
959 	default:
960 		return IORES_DESC_NONE;
961 	}
962 }
963 
964 static bool do_mark_busy(u32 type, struct resource *res)
965 {
966 	/* this is the legacy bios/dos rom-shadow + mmio region */
967 	if (res->start < (1ULL<<20))
968 		return true;
969 
970 	/*
971 	 * Treat persistent memory like device memory, i.e. reserve it
972 	 * for exclusive use of a driver
973 	 */
974 	switch (type) {
975 	case E820_RESERVED:
976 	case E820_PRAM:
977 	case E820_PMEM:
978 		return false;
979 	default:
980 		return true;
981 	}
982 }
983 
984 /*
985  * Mark e820 reserved areas as busy for the resource manager.
986  */
987 static struct resource __initdata *e820_res;
988 void __init e820_reserve_resources(void)
989 {
990 	int i;
991 	struct resource *res;
992 	u64 end;
993 
994 	res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
995 	e820_res = res;
996 	for (i = 0; i < e820.nr_map; i++) {
997 		end = e820.map[i].addr + e820.map[i].size - 1;
998 		if (end != (resource_size_t)end) {
999 			res++;
1000 			continue;
1001 		}
1002 		res->name = e820_type_to_string(e820.map[i].type);
1003 		res->start = e820.map[i].addr;
1004 		res->end = end;
1005 
1006 		res->flags = e820_type_to_iomem_type(e820.map[i].type);
1007 		res->desc = e820_type_to_iores_desc(e820.map[i].type);
1008 
1009 		/*
1010 		 * don't register the region that could be conflicted with
1011 		 * pci device BAR resource and insert them later in
1012 		 * pcibios_resource_survey()
1013 		 */
1014 		if (do_mark_busy(e820.map[i].type, res)) {
1015 			res->flags |= IORESOURCE_BUSY;
1016 			insert_resource(&iomem_resource, res);
1017 		}
1018 		res++;
1019 	}
1020 
1021 	for (i = 0; i < e820_saved.nr_map; i++) {
1022 		struct e820entry *entry = &e820_saved.map[i];
1023 		firmware_map_add_early(entry->addr,
1024 			entry->addr + entry->size,
1025 			e820_type_to_string(entry->type));
1026 	}
1027 }
1028 
1029 /* How much should we pad RAM ending depending on where it is? */
1030 static unsigned long ram_alignment(resource_size_t pos)
1031 {
1032 	unsigned long mb = pos >> 20;
1033 
1034 	/* To 64kB in the first megabyte */
1035 	if (!mb)
1036 		return 64*1024;
1037 
1038 	/* To 1MB in the first 16MB */
1039 	if (mb < 16)
1040 		return 1024*1024;
1041 
1042 	/* To 64MB for anything above that */
1043 	return 64*1024*1024;
1044 }
1045 
1046 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1047 
1048 void __init e820_reserve_resources_late(void)
1049 {
1050 	int i;
1051 	struct resource *res;
1052 
1053 	res = e820_res;
1054 	for (i = 0; i < e820.nr_map; i++) {
1055 		if (!res->parent && res->end)
1056 			insert_resource_expand_to_fit(&iomem_resource, res);
1057 		res++;
1058 	}
1059 
1060 	/*
1061 	 * Try to bump up RAM regions to reasonable boundaries to
1062 	 * avoid stolen RAM:
1063 	 */
1064 	for (i = 0; i < e820.nr_map; i++) {
1065 		struct e820entry *entry = &e820.map[i];
1066 		u64 start, end;
1067 
1068 		if (entry->type != E820_RAM)
1069 			continue;
1070 		start = entry->addr + entry->size;
1071 		end = round_up(start, ram_alignment(start)) - 1;
1072 		if (end > MAX_RESOURCE_SIZE)
1073 			end = MAX_RESOURCE_SIZE;
1074 		if (start >= end)
1075 			continue;
1076 		printk(KERN_DEBUG
1077 		       "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n",
1078 		       start, end);
1079 		reserve_region_with_split(&iomem_resource, start, end,
1080 					  "RAM buffer");
1081 	}
1082 }
1083 
1084 char *__init default_machine_specific_memory_setup(void)
1085 {
1086 	char *who = "BIOS-e820";
1087 	u32 new_nr;
1088 	/*
1089 	 * Try to copy the BIOS-supplied E820-map.
1090 	 *
1091 	 * Otherwise fake a memory map; one section from 0k->640k,
1092 	 * the next section from 1mb->appropriate_mem_k
1093 	 */
1094 	new_nr = boot_params.e820_entries;
1095 	sanitize_e820_map(boot_params.e820_map,
1096 			ARRAY_SIZE(boot_params.e820_map),
1097 			&new_nr);
1098 	boot_params.e820_entries = new_nr;
1099 	if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1100 	  < 0) {
1101 		u64 mem_size;
1102 
1103 		/* compare results from other methods and take the greater */
1104 		if (boot_params.alt_mem_k
1105 		    < boot_params.screen_info.ext_mem_k) {
1106 			mem_size = boot_params.screen_info.ext_mem_k;
1107 			who = "BIOS-88";
1108 		} else {
1109 			mem_size = boot_params.alt_mem_k;
1110 			who = "BIOS-e801";
1111 		}
1112 
1113 		e820.nr_map = 0;
1114 		e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1115 		e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1116 	}
1117 
1118 	/* In case someone cares... */
1119 	return who;
1120 }
1121 
1122 void __init setup_memory_map(void)
1123 {
1124 	char *who;
1125 
1126 	who = x86_init.resources.memory_setup();
1127 	memcpy(&e820_saved, &e820, sizeof(struct e820map));
1128 	printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n");
1129 	e820_print_map(who);
1130 }
1131 
1132 void __init memblock_x86_fill(void)
1133 {
1134 	int i;
1135 	u64 end;
1136 
1137 	/*
1138 	 * EFI may have more than 128 entries
1139 	 * We are safe to enable resizing, beause memblock_x86_fill()
1140 	 * is rather later for x86
1141 	 */
1142 	memblock_allow_resize();
1143 
1144 	for (i = 0; i < e820.nr_map; i++) {
1145 		struct e820entry *ei = &e820.map[i];
1146 
1147 		end = ei->addr + ei->size;
1148 		if (end != (resource_size_t)end)
1149 			continue;
1150 
1151 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1152 			continue;
1153 
1154 		memblock_add(ei->addr, ei->size);
1155 	}
1156 
1157 	/* throw away partial pages */
1158 	memblock_trim_memory(PAGE_SIZE);
1159 
1160 	memblock_dump_all();
1161 }
1162 
1163 void __init memblock_find_dma_reserve(void)
1164 {
1165 #ifdef CONFIG_X86_64
1166 	u64 nr_pages = 0, nr_free_pages = 0;
1167 	unsigned long start_pfn, end_pfn;
1168 	phys_addr_t start, end;
1169 	int i;
1170 	u64 u;
1171 
1172 	/*
1173 	 * need to find out used area below MAX_DMA_PFN
1174 	 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1175 	 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1176 	 */
1177 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1178 		start_pfn = min(start_pfn, MAX_DMA_PFN);
1179 		end_pfn = min(end_pfn, MAX_DMA_PFN);
1180 		nr_pages += end_pfn - start_pfn;
1181 	}
1182 
1183 	for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1184 				NULL) {
1185 		start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1186 		end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1187 		if (start_pfn < end_pfn)
1188 			nr_free_pages += end_pfn - start_pfn;
1189 	}
1190 
1191 	set_dma_reserve(nr_pages - nr_free_pages);
1192 #endif
1193 }
1194