xref: /linux/arch/x86/kernel/e820.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 	default:
153 		printk(KERN_CONT "type %u", type);
154 		break;
155 	}
156 }
157 
158 void __init e820_print_map(char *who)
159 {
160 	int i;
161 
162 	for (i = 0; i < e820.nr_map; i++) {
163 		printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who,
164 		       (unsigned long long) e820.map[i].addr,
165 		       (unsigned long long)
166 		       (e820.map[i].addr + e820.map[i].size - 1));
167 		e820_print_type(e820.map[i].type);
168 		printk(KERN_CONT "\n");
169 	}
170 }
171 
172 /*
173  * Sanitize the BIOS e820 map.
174  *
175  * Some e820 responses include overlapping entries. The following
176  * replaces the original e820 map with a new one, removing overlaps,
177  * and resolving conflicting memory types in favor of highest
178  * numbered type.
179  *
180  * The input parameter biosmap points to an array of 'struct
181  * e820entry' which on entry has elements in the range [0, *pnr_map)
182  * valid, and which has space for up to max_nr_map entries.
183  * On return, the resulting sanitized e820 map entries will be in
184  * overwritten in the same location, starting at biosmap.
185  *
186  * The integer pointed to by pnr_map must be valid on entry (the
187  * current number of valid entries located at biosmap) and will
188  * be updated on return, with the new number of valid entries
189  * (something no more than max_nr_map.)
190  *
191  * The return value from sanitize_e820_map() is zero if it
192  * successfully 'sanitized' the map entries passed in, and is -1
193  * if it did nothing, which can happen if either of (1) it was
194  * only passed one map entry, or (2) any of the input map entries
195  * were invalid (start + size < start, meaning that the size was
196  * so big the described memory range wrapped around through zero.)
197  *
198  *	Visually we're performing the following
199  *	(1,2,3,4 = memory types)...
200  *
201  *	Sample memory map (w/overlaps):
202  *	   ____22__________________
203  *	   ______________________4_
204  *	   ____1111________________
205  *	   _44_____________________
206  *	   11111111________________
207  *	   ____________________33__
208  *	   ___________44___________
209  *	   __________33333_________
210  *	   ______________22________
211  *	   ___________________2222_
212  *	   _________111111111______
213  *	   _____________________11_
214  *	   _________________4______
215  *
216  *	Sanitized equivalent (no overlap):
217  *	   1_______________________
218  *	   _44_____________________
219  *	   ___1____________________
220  *	   ____22__________________
221  *	   ______11________________
222  *	   _________1______________
223  *	   __________3_____________
224  *	   ___________44___________
225  *	   _____________33_________
226  *	   _______________2________
227  *	   ________________1_______
228  *	   _________________4______
229  *	   ___________________2____
230  *	   ____________________33__
231  *	   ______________________4_
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 
238 static int __init cpcompare(const void *a, const void *b)
239 {
240 	struct change_member * const *app = a, * const *bpp = b;
241 	const struct change_member *ap = *app, *bp = *bpp;
242 
243 	/*
244 	 * Inputs are pointers to two elements of change_point[].  If their
245 	 * addresses are unequal, their difference dominates.  If the addresses
246 	 * are equal, then consider one that represents the end of its region
247 	 * to be greater than one that does not.
248 	 */
249 	if (ap->addr != bp->addr)
250 		return ap->addr > bp->addr ? 1 : -1;
251 
252 	return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr);
253 }
254 
255 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
256 			     u32 *pnr_map)
257 {
258 	static struct change_member change_point_list[2*E820_X_MAX] __initdata;
259 	static struct change_member *change_point[2*E820_X_MAX] __initdata;
260 	static struct e820entry *overlap_list[E820_X_MAX] __initdata;
261 	static struct e820entry new_bios[E820_X_MAX] __initdata;
262 	unsigned long current_type, last_type;
263 	unsigned long long last_addr;
264 	int chgidx;
265 	int overlap_entries;
266 	int new_bios_entry;
267 	int old_nr, new_nr, chg_nr;
268 	int i;
269 
270 	/* if there's only one memory region, don't bother */
271 	if (*pnr_map < 2)
272 		return -1;
273 
274 	old_nr = *pnr_map;
275 	BUG_ON(old_nr > max_nr_map);
276 
277 	/* bail out if we find any unreasonable addresses in bios map */
278 	for (i = 0; i < old_nr; i++)
279 		if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
280 			return -1;
281 
282 	/* create pointers for initial change-point information (for sorting) */
283 	for (i = 0; i < 2 * old_nr; i++)
284 		change_point[i] = &change_point_list[i];
285 
286 	/* record all known change-points (starting and ending addresses),
287 	   omitting those that are for empty memory regions */
288 	chgidx = 0;
289 	for (i = 0; i < old_nr; i++)	{
290 		if (biosmap[i].size != 0) {
291 			change_point[chgidx]->addr = biosmap[i].addr;
292 			change_point[chgidx++]->pbios = &biosmap[i];
293 			change_point[chgidx]->addr = biosmap[i].addr +
294 				biosmap[i].size;
295 			change_point[chgidx++]->pbios = &biosmap[i];
296 		}
297 	}
298 	chg_nr = chgidx;
299 
300 	/* sort change-point list by memory addresses (low -> high) */
301 	sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL);
302 
303 	/* create a new bios memory map, removing overlaps */
304 	overlap_entries = 0;	 /* number of entries in the overlap table */
305 	new_bios_entry = 0;	 /* index for creating new bios map entries */
306 	last_type = 0;		 /* start with undefined memory type */
307 	last_addr = 0;		 /* start with 0 as last starting address */
308 
309 	/* loop through change-points, determining affect on the new bios map */
310 	for (chgidx = 0; chgidx < chg_nr; chgidx++) {
311 		/* keep track of all overlapping bios entries */
312 		if (change_point[chgidx]->addr ==
313 		    change_point[chgidx]->pbios->addr) {
314 			/*
315 			 * add map entry to overlap list (> 1 entry
316 			 * implies an overlap)
317 			 */
318 			overlap_list[overlap_entries++] =
319 				change_point[chgidx]->pbios;
320 		} else {
321 			/*
322 			 * remove entry from list (order independent,
323 			 * so swap with last)
324 			 */
325 			for (i = 0; i < overlap_entries; i++) {
326 				if (overlap_list[i] ==
327 				    change_point[chgidx]->pbios)
328 					overlap_list[i] =
329 						overlap_list[overlap_entries-1];
330 			}
331 			overlap_entries--;
332 		}
333 		/*
334 		 * if there are overlapping entries, decide which
335 		 * "type" to use (larger value takes precedence --
336 		 * 1=usable, 2,3,4,4+=unusable)
337 		 */
338 		current_type = 0;
339 		for (i = 0; i < overlap_entries; i++)
340 			if (overlap_list[i]->type > current_type)
341 				current_type = overlap_list[i]->type;
342 		/*
343 		 * continue building up new bios map based on this
344 		 * information
345 		 */
346 		if (current_type != last_type)	{
347 			if (last_type != 0)	 {
348 				new_bios[new_bios_entry].size =
349 					change_point[chgidx]->addr - last_addr;
350 				/*
351 				 * move forward only if the new size
352 				 * was non-zero
353 				 */
354 				if (new_bios[new_bios_entry].size != 0)
355 					/*
356 					 * no more space left for new
357 					 * bios entries ?
358 					 */
359 					if (++new_bios_entry >= max_nr_map)
360 						break;
361 			}
362 			if (current_type != 0)	{
363 				new_bios[new_bios_entry].addr =
364 					change_point[chgidx]->addr;
365 				new_bios[new_bios_entry].type = current_type;
366 				last_addr = change_point[chgidx]->addr;
367 			}
368 			last_type = current_type;
369 		}
370 	}
371 	/* retain count for new bios entries */
372 	new_nr = new_bios_entry;
373 
374 	/* copy new bios mapping into original location */
375 	memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
376 	*pnr_map = new_nr;
377 
378 	return 0;
379 }
380 
381 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
382 {
383 	while (nr_map) {
384 		u64 start = biosmap->addr;
385 		u64 size = biosmap->size;
386 		u64 end = start + size;
387 		u32 type = biosmap->type;
388 
389 		/* Overflow in 64 bits? Ignore the memory map. */
390 		if (start > end)
391 			return -1;
392 
393 		e820_add_region(start, size, type);
394 
395 		biosmap++;
396 		nr_map--;
397 	}
398 	return 0;
399 }
400 
401 /*
402  * Copy the BIOS e820 map into a safe place.
403  *
404  * Sanity-check it while we're at it..
405  *
406  * If we're lucky and live on a modern system, the setup code
407  * will have given us a memory map that we can use to properly
408  * set up memory.  If we aren't, we'll fake a memory map.
409  */
410 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
411 {
412 	/* Only one memory region (or negative)? Ignore it */
413 	if (nr_map < 2)
414 		return -1;
415 
416 	return __append_e820_map(biosmap, nr_map);
417 }
418 
419 static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
420 					u64 size, unsigned old_type,
421 					unsigned new_type)
422 {
423 	u64 end;
424 	unsigned int i;
425 	u64 real_updated_size = 0;
426 
427 	BUG_ON(old_type == new_type);
428 
429 	if (size > (ULLONG_MAX - start))
430 		size = ULLONG_MAX - start;
431 
432 	end = start + size;
433 	printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ",
434 	       (unsigned long long) start, (unsigned long long) (end - 1));
435 	e820_print_type(old_type);
436 	printk(KERN_CONT " ==> ");
437 	e820_print_type(new_type);
438 	printk(KERN_CONT "\n");
439 
440 	for (i = 0; i < e820x->nr_map; i++) {
441 		struct e820entry *ei = &e820x->map[i];
442 		u64 final_start, final_end;
443 		u64 ei_end;
444 
445 		if (ei->type != old_type)
446 			continue;
447 
448 		ei_end = ei->addr + ei->size;
449 		/* totally covered by new range? */
450 		if (ei->addr >= start && ei_end <= end) {
451 			ei->type = new_type;
452 			real_updated_size += ei->size;
453 			continue;
454 		}
455 
456 		/* new range is totally covered? */
457 		if (ei->addr < start && ei_end > end) {
458 			__e820_add_region(e820x, start, size, new_type);
459 			__e820_add_region(e820x, end, ei_end - end, ei->type);
460 			ei->size = start - ei->addr;
461 			real_updated_size += size;
462 			continue;
463 		}
464 
465 		/* partially covered */
466 		final_start = max(start, ei->addr);
467 		final_end = min(end, ei_end);
468 		if (final_start >= final_end)
469 			continue;
470 
471 		__e820_add_region(e820x, final_start, final_end - final_start,
472 				  new_type);
473 
474 		real_updated_size += final_end - final_start;
475 
476 		/*
477 		 * left range could be head or tail, so need to update
478 		 * size at first.
479 		 */
480 		ei->size -= final_end - final_start;
481 		if (ei->addr < final_start)
482 			continue;
483 		ei->addr = final_end;
484 	}
485 	return real_updated_size;
486 }
487 
488 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
489 			     unsigned new_type)
490 {
491 	return __e820_update_range(&e820, start, size, old_type, new_type);
492 }
493 
494 static u64 __init e820_update_range_saved(u64 start, u64 size,
495 					  unsigned old_type, unsigned new_type)
496 {
497 	return __e820_update_range(&e820_saved, start, size, old_type,
498 				     new_type);
499 }
500 
501 /* make e820 not cover the range */
502 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
503 			     int checktype)
504 {
505 	int i;
506 	u64 end;
507 	u64 real_removed_size = 0;
508 
509 	if (size > (ULLONG_MAX - start))
510 		size = ULLONG_MAX - start;
511 
512 	end = start + size;
513 	printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ",
514 	       (unsigned long long) start, (unsigned long long) (end - 1));
515 	if (checktype)
516 		e820_print_type(old_type);
517 	printk(KERN_CONT "\n");
518 
519 	for (i = 0; i < e820.nr_map; i++) {
520 		struct e820entry *ei = &e820.map[i];
521 		u64 final_start, final_end;
522 		u64 ei_end;
523 
524 		if (checktype && ei->type != old_type)
525 			continue;
526 
527 		ei_end = ei->addr + ei->size;
528 		/* totally covered? */
529 		if (ei->addr >= start && ei_end <= end) {
530 			real_removed_size += ei->size;
531 			memset(ei, 0, sizeof(struct e820entry));
532 			continue;
533 		}
534 
535 		/* new range is totally covered? */
536 		if (ei->addr < start && ei_end > end) {
537 			e820_add_region(end, ei_end - end, ei->type);
538 			ei->size = start - ei->addr;
539 			real_removed_size += size;
540 			continue;
541 		}
542 
543 		/* partially covered */
544 		final_start = max(start, ei->addr);
545 		final_end = min(end, ei_end);
546 		if (final_start >= final_end)
547 			continue;
548 		real_removed_size += final_end - final_start;
549 
550 		/*
551 		 * left range could be head or tail, so need to update
552 		 * size at first.
553 		 */
554 		ei->size -= final_end - final_start;
555 		if (ei->addr < final_start)
556 			continue;
557 		ei->addr = final_end;
558 	}
559 	return real_removed_size;
560 }
561 
562 void __init update_e820(void)
563 {
564 	u32 nr_map;
565 
566 	nr_map = e820.nr_map;
567 	if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
568 		return;
569 	e820.nr_map = nr_map;
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 	u32 nr_map;
576 
577 	nr_map = e820_saved.nr_map;
578 	if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
579 		return;
580 	e820_saved.nr_map = nr_map;
581 }
582 #define MAX_GAP_END 0x100000000ull
583 /*
584  * Search for a gap in the e820 memory space from start_addr to end_addr.
585  */
586 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
587 		unsigned long start_addr, unsigned long long end_addr)
588 {
589 	unsigned long long last;
590 	int i = e820.nr_map;
591 	int found = 0;
592 
593 	last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
594 
595 	while (--i >= 0) {
596 		unsigned long long start = e820.map[i].addr;
597 		unsigned long long end = start + e820.map[i].size;
598 
599 		if (end < start_addr)
600 			continue;
601 
602 		/*
603 		 * Since "last" is at most 4GB, we know we'll
604 		 * fit in 32 bits if this condition is true
605 		 */
606 		if (last > end) {
607 			unsigned long gap = last - end;
608 
609 			if (gap >= *gapsize) {
610 				*gapsize = gap;
611 				*gapstart = end;
612 				found = 1;
613 			}
614 		}
615 		if (start < last)
616 			last = start;
617 	}
618 	return found;
619 }
620 
621 /*
622  * Search for the biggest gap in the low 32 bits of the e820
623  * memory space.  We pass this space to PCI to assign MMIO resources
624  * for hotplug or unconfigured devices in.
625  * Hopefully the BIOS let enough space left.
626  */
627 __init void e820_setup_gap(void)
628 {
629 	unsigned long gapstart, gapsize;
630 	int found;
631 
632 	gapstart = 0x10000000;
633 	gapsize = 0x400000;
634 	found  = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
635 
636 #ifdef CONFIG_X86_64
637 	if (!found) {
638 		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
639 		printk(KERN_ERR
640 	"e820: cannot find a gap in the 32bit address range\n"
641 	"e820: PCI devices with unassigned 32bit BARs may break!\n");
642 	}
643 #endif
644 
645 	/*
646 	 * e820_reserve_resources_late protect stolen RAM already
647 	 */
648 	pci_mem_start = gapstart;
649 
650 	printk(KERN_INFO
651 	       "e820: [mem %#010lx-%#010lx] available for PCI devices\n",
652 	       gapstart, gapstart + gapsize - 1);
653 }
654 
655 /**
656  * Because of the size limitation of struct boot_params, only first
657  * 128 E820 memory entries are passed to kernel via
658  * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
659  * linked list of struct setup_data, which is parsed here.
660  */
661 void __init parse_e820_ext(struct setup_data *sdata)
662 {
663 	int entries;
664 	struct e820entry *extmap;
665 
666 	entries = sdata->len / sizeof(struct e820entry);
667 	extmap = (struct e820entry *)(sdata->data);
668 	__append_e820_map(extmap, entries);
669 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
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 and assumes the first e820 area to be RAM.
683  */
684 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
685 {
686 	int i;
687 	unsigned long pfn;
688 
689 	pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
690 	for (i = 1; i < e820.nr_map; i++) {
691 		struct e820entry *ei = &e820.map[i];
692 
693 		if (pfn < PFN_UP(ei->addr))
694 			register_nosave_region(pfn, PFN_UP(ei->addr));
695 
696 		pfn = PFN_DOWN(ei->addr + ei->size);
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, unsigned type)
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 		if (ei->type != type)
769 			continue;
770 
771 		start_pfn = ei->addr >> PAGE_SHIFT;
772 		end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
773 
774 		if (start_pfn >= limit_pfn)
775 			continue;
776 		if (end_pfn > limit_pfn) {
777 			last_pfn = limit_pfn;
778 			break;
779 		}
780 		if (end_pfn > last_pfn)
781 			last_pfn = end_pfn;
782 	}
783 
784 	if (last_pfn > max_arch_pfn)
785 		last_pfn = max_arch_pfn;
786 
787 	printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
788 			 last_pfn, max_arch_pfn);
789 	return last_pfn;
790 }
791 unsigned long __init e820_end_of_ram_pfn(void)
792 {
793 	return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
794 }
795 
796 unsigned long __init e820_end_of_low_ram_pfn(void)
797 {
798 	return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
799 }
800 
801 static void early_panic(char *msg)
802 {
803 	early_printk(msg);
804 	panic(msg);
805 }
806 
807 static int userdef __initdata;
808 
809 /* "mem=nopentium" disables the 4MB page tables. */
810 static int __init parse_memopt(char *p)
811 {
812 	u64 mem_size;
813 
814 	if (!p)
815 		return -EINVAL;
816 
817 	if (!strcmp(p, "nopentium")) {
818 #ifdef CONFIG_X86_32
819 		setup_clear_cpu_cap(X86_FEATURE_PSE);
820 		return 0;
821 #else
822 		printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
823 		return -EINVAL;
824 #endif
825 	}
826 
827 	userdef = 1;
828 	mem_size = memparse(p, &p);
829 	/* don't remove all of memory when handling "mem={invalid}" param */
830 	if (mem_size == 0)
831 		return -EINVAL;
832 	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
833 
834 	return 0;
835 }
836 early_param("mem", parse_memopt);
837 
838 static int __init parse_memmap_opt(char *p)
839 {
840 	char *oldp;
841 	u64 start_at, mem_size;
842 
843 	if (!p)
844 		return -EINVAL;
845 
846 	if (!strncmp(p, "exactmap", 8)) {
847 #ifdef CONFIG_CRASH_DUMP
848 		/*
849 		 * If we are doing a crash dump, we still need to know
850 		 * the real mem size before original memory map is
851 		 * reset.
852 		 */
853 		saved_max_pfn = e820_end_of_ram_pfn();
854 #endif
855 		e820.nr_map = 0;
856 		userdef = 1;
857 		return 0;
858 	}
859 
860 	oldp = p;
861 	mem_size = memparse(p, &p);
862 	if (p == oldp)
863 		return -EINVAL;
864 
865 	userdef = 1;
866 	if (*p == '@') {
867 		start_at = memparse(p+1, &p);
868 		e820_add_region(start_at, mem_size, E820_RAM);
869 	} else if (*p == '#') {
870 		start_at = memparse(p+1, &p);
871 		e820_add_region(start_at, mem_size, E820_ACPI);
872 	} else if (*p == '$') {
873 		start_at = memparse(p+1, &p);
874 		e820_add_region(start_at, mem_size, E820_RESERVED);
875 	} else
876 		e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
877 
878 	return *p == '\0' ? 0 : -EINVAL;
879 }
880 early_param("memmap", parse_memmap_opt);
881 
882 void __init finish_e820_parsing(void)
883 {
884 	if (userdef) {
885 		u32 nr = e820.nr_map;
886 
887 		if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
888 			early_panic("Invalid user supplied memory map");
889 		e820.nr_map = nr;
890 
891 		printk(KERN_INFO "e820: user-defined physical RAM map:\n");
892 		e820_print_map("user");
893 	}
894 }
895 
896 static inline const char *e820_type_to_string(int e820_type)
897 {
898 	switch (e820_type) {
899 	case E820_RESERVED_KERN:
900 	case E820_RAM:	return "System RAM";
901 	case E820_ACPI:	return "ACPI Tables";
902 	case E820_NVS:	return "ACPI Non-volatile Storage";
903 	case E820_UNUSABLE:	return "Unusable memory";
904 	default:	return "reserved";
905 	}
906 }
907 
908 /*
909  * Mark e820 reserved areas as busy for the resource manager.
910  */
911 static struct resource __initdata *e820_res;
912 void __init e820_reserve_resources(void)
913 {
914 	int i;
915 	struct resource *res;
916 	u64 end;
917 
918 	res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
919 	e820_res = res;
920 	for (i = 0; i < e820.nr_map; i++) {
921 		end = e820.map[i].addr + e820.map[i].size - 1;
922 		if (end != (resource_size_t)end) {
923 			res++;
924 			continue;
925 		}
926 		res->name = e820_type_to_string(e820.map[i].type);
927 		res->start = e820.map[i].addr;
928 		res->end = end;
929 
930 		res->flags = IORESOURCE_MEM;
931 
932 		/*
933 		 * don't register the region that could be conflicted with
934 		 * pci device BAR resource and insert them later in
935 		 * pcibios_resource_survey()
936 		 */
937 		if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20)) {
938 			res->flags |= IORESOURCE_BUSY;
939 			insert_resource(&iomem_resource, res);
940 		}
941 		res++;
942 	}
943 
944 	for (i = 0; i < e820_saved.nr_map; i++) {
945 		struct e820entry *entry = &e820_saved.map[i];
946 		firmware_map_add_early(entry->addr,
947 			entry->addr + entry->size,
948 			e820_type_to_string(entry->type));
949 	}
950 }
951 
952 /* How much should we pad RAM ending depending on where it is? */
953 static unsigned long ram_alignment(resource_size_t pos)
954 {
955 	unsigned long mb = pos >> 20;
956 
957 	/* To 64kB in the first megabyte */
958 	if (!mb)
959 		return 64*1024;
960 
961 	/* To 1MB in the first 16MB */
962 	if (mb < 16)
963 		return 1024*1024;
964 
965 	/* To 64MB for anything above that */
966 	return 64*1024*1024;
967 }
968 
969 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
970 
971 void __init e820_reserve_resources_late(void)
972 {
973 	int i;
974 	struct resource *res;
975 
976 	res = e820_res;
977 	for (i = 0; i < e820.nr_map; i++) {
978 		if (!res->parent && res->end)
979 			insert_resource_expand_to_fit(&iomem_resource, res);
980 		res++;
981 	}
982 
983 	/*
984 	 * Try to bump up RAM regions to reasonable boundaries to
985 	 * avoid stolen RAM:
986 	 */
987 	for (i = 0; i < e820.nr_map; i++) {
988 		struct e820entry *entry = &e820.map[i];
989 		u64 start, end;
990 
991 		if (entry->type != E820_RAM)
992 			continue;
993 		start = entry->addr + entry->size;
994 		end = round_up(start, ram_alignment(start)) - 1;
995 		if (end > MAX_RESOURCE_SIZE)
996 			end = MAX_RESOURCE_SIZE;
997 		if (start >= end)
998 			continue;
999 		printk(KERN_DEBUG
1000 		       "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n",
1001 		       start, end);
1002 		reserve_region_with_split(&iomem_resource, start, end,
1003 					  "RAM buffer");
1004 	}
1005 }
1006 
1007 char *__init default_machine_specific_memory_setup(void)
1008 {
1009 	char *who = "BIOS-e820";
1010 	u32 new_nr;
1011 	/*
1012 	 * Try to copy the BIOS-supplied E820-map.
1013 	 *
1014 	 * Otherwise fake a memory map; one section from 0k->640k,
1015 	 * the next section from 1mb->appropriate_mem_k
1016 	 */
1017 	new_nr = boot_params.e820_entries;
1018 	sanitize_e820_map(boot_params.e820_map,
1019 			ARRAY_SIZE(boot_params.e820_map),
1020 			&new_nr);
1021 	boot_params.e820_entries = new_nr;
1022 	if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1023 	  < 0) {
1024 		u64 mem_size;
1025 
1026 		/* compare results from other methods and take the greater */
1027 		if (boot_params.alt_mem_k
1028 		    < boot_params.screen_info.ext_mem_k) {
1029 			mem_size = boot_params.screen_info.ext_mem_k;
1030 			who = "BIOS-88";
1031 		} else {
1032 			mem_size = boot_params.alt_mem_k;
1033 			who = "BIOS-e801";
1034 		}
1035 
1036 		e820.nr_map = 0;
1037 		e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1038 		e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1039 	}
1040 
1041 	/* In case someone cares... */
1042 	return who;
1043 }
1044 
1045 void __init setup_memory_map(void)
1046 {
1047 	char *who;
1048 
1049 	who = x86_init.resources.memory_setup();
1050 	memcpy(&e820_saved, &e820, sizeof(struct e820map));
1051 	printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n");
1052 	e820_print_map(who);
1053 }
1054 
1055 void __init memblock_x86_fill(void)
1056 {
1057 	int i;
1058 	u64 end;
1059 
1060 	/*
1061 	 * EFI may have more than 128 entries
1062 	 * We are safe to enable resizing, beause memblock_x86_fill()
1063 	 * is rather later for x86
1064 	 */
1065 	memblock_allow_resize();
1066 
1067 	for (i = 0; i < e820.nr_map; i++) {
1068 		struct e820entry *ei = &e820.map[i];
1069 
1070 		end = ei->addr + ei->size;
1071 		if (end != (resource_size_t)end)
1072 			continue;
1073 
1074 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1075 			continue;
1076 
1077 		memblock_add(ei->addr, ei->size);
1078 	}
1079 
1080 	/* throw away partial pages */
1081 	memblock_trim_memory(PAGE_SIZE);
1082 
1083 	memblock_dump_all();
1084 }
1085 
1086 void __init memblock_find_dma_reserve(void)
1087 {
1088 #ifdef CONFIG_X86_64
1089 	u64 nr_pages = 0, nr_free_pages = 0;
1090 	unsigned long start_pfn, end_pfn;
1091 	phys_addr_t start, end;
1092 	int i;
1093 	u64 u;
1094 
1095 	/*
1096 	 * need to find out used area below MAX_DMA_PFN
1097 	 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1098 	 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1099 	 */
1100 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1101 		start_pfn = min_t(unsigned long, start_pfn, MAX_DMA_PFN);
1102 		end_pfn = min_t(unsigned long, end_pfn, MAX_DMA_PFN);
1103 		nr_pages += end_pfn - start_pfn;
1104 	}
1105 
1106 	for_each_free_mem_range(u, MAX_NUMNODES, &start, &end, NULL) {
1107 		start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1108 		end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1109 		if (start_pfn < end_pfn)
1110 			nr_free_pages += end_pfn - start_pfn;
1111 	}
1112 
1113 	set_dma_reserve(nr_pages - nr_free_pages);
1114 #endif
1115 }
1116