xref: /linux/arch/x86/kernel/e820.c (revision a3a4a816b4b194c45d0217e8b9e08b2639802cda)
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 static struct e820map initial_e820  __initdata;
44 static struct e820map initial_e820_saved  __initdata;
45 struct e820map *e820 __refdata = &initial_e820;
46 struct e820map *e820_saved __refdata = &initial_e820_saved;
47 
48 /* For PCI or other memory-mapped resources */
49 unsigned long pci_mem_start = 0xaeedbabe;
50 #ifdef CONFIG_PCI
51 EXPORT_SYMBOL(pci_mem_start);
52 #endif
53 
54 /*
55  * This function checks if any part of the range <start,end> is mapped
56  * with type.
57  */
58 int
59 e820_any_mapped(u64 start, u64 end, unsigned type)
60 {
61 	int i;
62 
63 	for (i = 0; i < e820->nr_map; i++) {
64 		struct e820entry *ei = &e820->map[i];
65 
66 		if (type && ei->type != type)
67 			continue;
68 		if (ei->addr >= end || ei->addr + ei->size <= start)
69 			continue;
70 		return 1;
71 	}
72 	return 0;
73 }
74 EXPORT_SYMBOL_GPL(e820_any_mapped);
75 
76 /*
77  * This function checks if the entire range <start,end> is mapped with type.
78  *
79  * Note: this function only works correct if the e820 table is sorted and
80  * not-overlapping, which is the case
81  */
82 int __init e820_all_mapped(u64 start, u64 end, unsigned type)
83 {
84 	int i;
85 
86 	for (i = 0; i < e820->nr_map; i++) {
87 		struct e820entry *ei = &e820->map[i];
88 
89 		if (type && ei->type != type)
90 			continue;
91 		/* is the region (part) in overlap with the current region ?*/
92 		if (ei->addr >= end || ei->addr + ei->size <= start)
93 			continue;
94 
95 		/* if the region is at the beginning of <start,end> we move
96 		 * start to the end of the region since it's ok until there
97 		 */
98 		if (ei->addr <= start)
99 			start = ei->addr + ei->size;
100 		/*
101 		 * if start is now at or beyond end, we're done, full
102 		 * coverage
103 		 */
104 		if (start >= end)
105 			return 1;
106 	}
107 	return 0;
108 }
109 
110 /*
111  * Add a memory region to the kernel e820 map.
112  */
113 static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
114 					 int type)
115 {
116 	int x = e820x->nr_map;
117 
118 	if (x >= ARRAY_SIZE(e820x->map)) {
119 		printk(KERN_ERR "e820: too many entries; ignoring [mem %#010llx-%#010llx]\n",
120 		       (unsigned long long) start,
121 		       (unsigned long long) (start + size - 1));
122 		return;
123 	}
124 
125 	e820x->map[x].addr = start;
126 	e820x->map[x].size = size;
127 	e820x->map[x].type = type;
128 	e820x->nr_map++;
129 }
130 
131 void __init e820_add_region(u64 start, u64 size, int type)
132 {
133 	__e820_add_region(e820, start, size, type);
134 }
135 
136 static void __init e820_print_type(u32 type)
137 {
138 	switch (type) {
139 	case E820_RAM:
140 	case E820_RESERVED_KERN:
141 		printk(KERN_CONT "usable");
142 		break;
143 	case E820_RESERVED:
144 		printk(KERN_CONT "reserved");
145 		break;
146 	case E820_ACPI:
147 		printk(KERN_CONT "ACPI data");
148 		break;
149 	case E820_NVS:
150 		printk(KERN_CONT "ACPI NVS");
151 		break;
152 	case E820_UNUSABLE:
153 		printk(KERN_CONT "unusable");
154 		break;
155 	case E820_PMEM:
156 	case E820_PRAM:
157 		printk(KERN_CONT "persistent (type %u)", type);
158 		break;
159 	default:
160 		printk(KERN_CONT "type %u", type);
161 		break;
162 	}
163 }
164 
165 void __init e820_print_map(char *who)
166 {
167 	int i;
168 
169 	for (i = 0; i < e820->nr_map; i++) {
170 		printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who,
171 		       (unsigned long long) e820->map[i].addr,
172 		       (unsigned long long)
173 		       (e820->map[i].addr + e820->map[i].size - 1));
174 		e820_print_type(e820->map[i].type);
175 		printk(KERN_CONT "\n");
176 	}
177 }
178 
179 /*
180  * Sanitize the BIOS e820 map.
181  *
182  * Some e820 responses include overlapping entries. The following
183  * replaces the original e820 map with a new one, removing overlaps,
184  * and resolving conflicting memory types in favor of highest
185  * numbered type.
186  *
187  * The input parameter biosmap points to an array of 'struct
188  * e820entry' which on entry has elements in the range [0, *pnr_map)
189  * valid, and which has space for up to max_nr_map entries.
190  * On return, the resulting sanitized e820 map entries will be in
191  * overwritten in the same location, starting at biosmap.
192  *
193  * The integer pointed to by pnr_map must be valid on entry (the
194  * current number of valid entries located at biosmap). If the
195  * sanitizing succeeds the *pnr_map will be updated with the new
196  * number of valid entries (something no more than max_nr_map).
197  *
198  * The return value from sanitize_e820_map() is zero if it
199  * successfully 'sanitized' the map entries passed in, and is -1
200  * if it did nothing, which can happen if either of (1) it was
201  * only passed one map entry, or (2) any of the input map entries
202  * were invalid (start + size < start, meaning that the size was
203  * so big the described memory range wrapped around through zero.)
204  *
205  *	Visually we're performing the following
206  *	(1,2,3,4 = memory types)...
207  *
208  *	Sample memory map (w/overlaps):
209  *	   ____22__________________
210  *	   ______________________4_
211  *	   ____1111________________
212  *	   _44_____________________
213  *	   11111111________________
214  *	   ____________________33__
215  *	   ___________44___________
216  *	   __________33333_________
217  *	   ______________22________
218  *	   ___________________2222_
219  *	   _________111111111______
220  *	   _____________________11_
221  *	   _________________4______
222  *
223  *	Sanitized equivalent (no overlap):
224  *	   1_______________________
225  *	   _44_____________________
226  *	   ___1____________________
227  *	   ____22__________________
228  *	   ______11________________
229  *	   _________1______________
230  *	   __________3_____________
231  *	   ___________44___________
232  *	   _____________33_________
233  *	   _______________2________
234  *	   ________________1_______
235  *	   _________________4______
236  *	   ___________________2____
237  *	   ____________________33__
238  *	   ______________________4_
239  */
240 struct change_member {
241 	struct e820entry *pbios; /* pointer to original bios entry */
242 	unsigned long long addr; /* address for this change point */
243 };
244 
245 static int __init cpcompare(const void *a, const void *b)
246 {
247 	struct change_member * const *app = a, * const *bpp = b;
248 	const struct change_member *ap = *app, *bp = *bpp;
249 
250 	/*
251 	 * Inputs are pointers to two elements of change_point[].  If their
252 	 * addresses are unequal, their difference dominates.  If the addresses
253 	 * are equal, then consider one that represents the end of its region
254 	 * to be greater than one that does not.
255 	 */
256 	if (ap->addr != bp->addr)
257 		return ap->addr > bp->addr ? 1 : -1;
258 
259 	return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr);
260 }
261 
262 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
263 			     u32 *pnr_map)
264 {
265 	static struct change_member change_point_list[2*E820_X_MAX] __initdata;
266 	static struct change_member *change_point[2*E820_X_MAX] __initdata;
267 	static struct e820entry *overlap_list[E820_X_MAX] __initdata;
268 	static struct e820entry new_bios[E820_X_MAX] __initdata;
269 	unsigned long current_type, last_type;
270 	unsigned long long last_addr;
271 	int chgidx;
272 	int overlap_entries;
273 	int new_bios_entry;
274 	int old_nr, new_nr, chg_nr;
275 	int i;
276 
277 	/* if there's only one memory region, don't bother */
278 	if (*pnr_map < 2)
279 		return -1;
280 
281 	old_nr = *pnr_map;
282 	BUG_ON(old_nr > max_nr_map);
283 
284 	/* bail out if we find any unreasonable addresses in bios map */
285 	for (i = 0; i < old_nr; i++)
286 		if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
287 			return -1;
288 
289 	/* create pointers for initial change-point information (for sorting) */
290 	for (i = 0; i < 2 * old_nr; i++)
291 		change_point[i] = &change_point_list[i];
292 
293 	/* record all known change-points (starting and ending addresses),
294 	   omitting those that are for empty memory regions */
295 	chgidx = 0;
296 	for (i = 0; i < old_nr; i++)	{
297 		if (biosmap[i].size != 0) {
298 			change_point[chgidx]->addr = biosmap[i].addr;
299 			change_point[chgidx++]->pbios = &biosmap[i];
300 			change_point[chgidx]->addr = biosmap[i].addr +
301 				biosmap[i].size;
302 			change_point[chgidx++]->pbios = &biosmap[i];
303 		}
304 	}
305 	chg_nr = chgidx;
306 
307 	/* sort change-point list by memory addresses (low -> high) */
308 	sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL);
309 
310 	/* create a new bios memory map, removing overlaps */
311 	overlap_entries = 0;	 /* number of entries in the overlap table */
312 	new_bios_entry = 0;	 /* index for creating new bios map entries */
313 	last_type = 0;		 /* start with undefined memory type */
314 	last_addr = 0;		 /* start with 0 as last starting address */
315 
316 	/* loop through change-points, determining affect on the new bios map */
317 	for (chgidx = 0; chgidx < chg_nr; chgidx++) {
318 		/* keep track of all overlapping bios entries */
319 		if (change_point[chgidx]->addr ==
320 		    change_point[chgidx]->pbios->addr) {
321 			/*
322 			 * add map entry to overlap list (> 1 entry
323 			 * implies an overlap)
324 			 */
325 			overlap_list[overlap_entries++] =
326 				change_point[chgidx]->pbios;
327 		} else {
328 			/*
329 			 * remove entry from list (order independent,
330 			 * so swap with last)
331 			 */
332 			for (i = 0; i < overlap_entries; i++) {
333 				if (overlap_list[i] ==
334 				    change_point[chgidx]->pbios)
335 					overlap_list[i] =
336 						overlap_list[overlap_entries-1];
337 			}
338 			overlap_entries--;
339 		}
340 		/*
341 		 * if there are overlapping entries, decide which
342 		 * "type" to use (larger value takes precedence --
343 		 * 1=usable, 2,3,4,4+=unusable)
344 		 */
345 		current_type = 0;
346 		for (i = 0; i < overlap_entries; i++)
347 			if (overlap_list[i]->type > current_type)
348 				current_type = overlap_list[i]->type;
349 		/*
350 		 * continue building up new bios map based on this
351 		 * information
352 		 */
353 		if (current_type != last_type || current_type == E820_PRAM) {
354 			if (last_type != 0)	 {
355 				new_bios[new_bios_entry].size =
356 					change_point[chgidx]->addr - last_addr;
357 				/*
358 				 * move forward only if the new size
359 				 * was non-zero
360 				 */
361 				if (new_bios[new_bios_entry].size != 0)
362 					/*
363 					 * no more space left for new
364 					 * bios entries ?
365 					 */
366 					if (++new_bios_entry >= max_nr_map)
367 						break;
368 			}
369 			if (current_type != 0)	{
370 				new_bios[new_bios_entry].addr =
371 					change_point[chgidx]->addr;
372 				new_bios[new_bios_entry].type = current_type;
373 				last_addr = change_point[chgidx]->addr;
374 			}
375 			last_type = current_type;
376 		}
377 	}
378 	/* retain count for new bios entries */
379 	new_nr = new_bios_entry;
380 
381 	/* copy new bios mapping into original location */
382 	memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
383 	*pnr_map = new_nr;
384 
385 	return 0;
386 }
387 
388 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
389 {
390 	while (nr_map) {
391 		u64 start = biosmap->addr;
392 		u64 size = biosmap->size;
393 		u64 end = start + size - 1;
394 		u32 type = biosmap->type;
395 
396 		/* Overflow in 64 bits? Ignore the memory map. */
397 		if (start > end && likely(size))
398 			return -1;
399 
400 		e820_add_region(start, size, type);
401 
402 		biosmap++;
403 		nr_map--;
404 	}
405 	return 0;
406 }
407 
408 /*
409  * Copy the BIOS e820 map into a safe place.
410  *
411  * Sanity-check it while we're at it..
412  *
413  * If we're lucky and live on a modern system, the setup code
414  * will have given us a memory map that we can use to properly
415  * set up memory.  If we aren't, we'll fake a memory map.
416  */
417 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
418 {
419 	/* Only one memory region (or negative)? Ignore it */
420 	if (nr_map < 2)
421 		return -1;
422 
423 	return __append_e820_map(biosmap, nr_map);
424 }
425 
426 static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
427 					u64 size, unsigned old_type,
428 					unsigned new_type)
429 {
430 	u64 end;
431 	unsigned int i;
432 	u64 real_updated_size = 0;
433 
434 	BUG_ON(old_type == new_type);
435 
436 	if (size > (ULLONG_MAX - start))
437 		size = ULLONG_MAX - start;
438 
439 	end = start + size;
440 	printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ",
441 	       (unsigned long long) start, (unsigned long long) (end - 1));
442 	e820_print_type(old_type);
443 	printk(KERN_CONT " ==> ");
444 	e820_print_type(new_type);
445 	printk(KERN_CONT "\n");
446 
447 	for (i = 0; i < e820x->nr_map; i++) {
448 		struct e820entry *ei = &e820x->map[i];
449 		u64 final_start, final_end;
450 		u64 ei_end;
451 
452 		if (ei->type != old_type)
453 			continue;
454 
455 		ei_end = ei->addr + ei->size;
456 		/* totally covered by new range? */
457 		if (ei->addr >= start && ei_end <= end) {
458 			ei->type = new_type;
459 			real_updated_size += ei->size;
460 			continue;
461 		}
462 
463 		/* new range is totally covered? */
464 		if (ei->addr < start && ei_end > end) {
465 			__e820_add_region(e820x, start, size, new_type);
466 			__e820_add_region(e820x, end, ei_end - end, ei->type);
467 			ei->size = start - ei->addr;
468 			real_updated_size += size;
469 			continue;
470 		}
471 
472 		/* partially covered */
473 		final_start = max(start, ei->addr);
474 		final_end = min(end, ei_end);
475 		if (final_start >= final_end)
476 			continue;
477 
478 		__e820_add_region(e820x, final_start, final_end - final_start,
479 				  new_type);
480 
481 		real_updated_size += final_end - final_start;
482 
483 		/*
484 		 * left range could be head or tail, so need to update
485 		 * size at first.
486 		 */
487 		ei->size -= final_end - final_start;
488 		if (ei->addr < final_start)
489 			continue;
490 		ei->addr = final_end;
491 	}
492 	return real_updated_size;
493 }
494 
495 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
496 			     unsigned new_type)
497 {
498 	return __e820_update_range(e820, start, size, old_type, new_type);
499 }
500 
501 static u64 __init e820_update_range_saved(u64 start, u64 size,
502 					  unsigned old_type, unsigned new_type)
503 {
504 	return __e820_update_range(e820_saved, start, size, old_type,
505 				     new_type);
506 }
507 
508 /* make e820 not cover the range */
509 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
510 			     int checktype)
511 {
512 	int i;
513 	u64 end;
514 	u64 real_removed_size = 0;
515 
516 	if (size > (ULLONG_MAX - start))
517 		size = ULLONG_MAX - start;
518 
519 	end = start + size;
520 	printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ",
521 	       (unsigned long long) start, (unsigned long long) (end - 1));
522 	if (checktype)
523 		e820_print_type(old_type);
524 	printk(KERN_CONT "\n");
525 
526 	for (i = 0; i < e820->nr_map; i++) {
527 		struct e820entry *ei = &e820->map[i];
528 		u64 final_start, final_end;
529 		u64 ei_end;
530 
531 		if (checktype && ei->type != old_type)
532 			continue;
533 
534 		ei_end = ei->addr + ei->size;
535 		/* totally covered? */
536 		if (ei->addr >= start && ei_end <= end) {
537 			real_removed_size += ei->size;
538 			memset(ei, 0, sizeof(struct e820entry));
539 			continue;
540 		}
541 
542 		/* new range is totally covered? */
543 		if (ei->addr < start && ei_end > end) {
544 			e820_add_region(end, ei_end - end, ei->type);
545 			ei->size = start - ei->addr;
546 			real_removed_size += size;
547 			continue;
548 		}
549 
550 		/* partially covered */
551 		final_start = max(start, ei->addr);
552 		final_end = min(end, ei_end);
553 		if (final_start >= final_end)
554 			continue;
555 		real_removed_size += final_end - final_start;
556 
557 		/*
558 		 * left range could be head or tail, so need to update
559 		 * size at first.
560 		 */
561 		ei->size -= final_end - final_start;
562 		if (ei->addr < final_start)
563 			continue;
564 		ei->addr = final_end;
565 	}
566 	return real_removed_size;
567 }
568 
569 void __init update_e820(void)
570 {
571 	if (sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map))
572 		return;
573 	printk(KERN_INFO "e820: modified physical RAM map:\n");
574 	e820_print_map("modified");
575 }
576 static void __init update_e820_saved(void)
577 {
578 	sanitize_e820_map(e820_saved->map, ARRAY_SIZE(e820_saved->map),
579 				&e820_saved->nr_map);
580 }
581 #define MAX_GAP_END 0x100000000ull
582 /*
583  * Search for a gap in the e820 memory space from 0 to MAX_GAP_END.
584  */
585 static int __init e820_search_gap(unsigned long *gapstart,
586 		unsigned long *gapsize)
587 {
588 	unsigned long long last = MAX_GAP_END;
589 	int i = e820->nr_map;
590 	int found = 0;
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 		/*
597 		 * Since "last" is at most 4GB, we know we'll
598 		 * fit in 32 bits if this condition is true
599 		 */
600 		if (last > end) {
601 			unsigned long gap = last - end;
602 
603 			if (gap >= *gapsize) {
604 				*gapsize = gap;
605 				*gapstart = end;
606 				found = 1;
607 			}
608 		}
609 		if (start < last)
610 			last = start;
611 	}
612 	return found;
613 }
614 
615 /*
616  * Search for the biggest gap in the low 32 bits of the e820
617  * memory space.  We pass this space to PCI to assign MMIO resources
618  * for hotplug or unconfigured devices in.
619  * Hopefully the BIOS let enough space left.
620  */
621 __init void e820_setup_gap(void)
622 {
623 	unsigned long gapstart, gapsize;
624 	int found;
625 
626 	gapsize = 0x400000;
627 	found  = e820_search_gap(&gapstart, &gapsize);
628 
629 	if (!found) {
630 #ifdef CONFIG_X86_64
631 		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
632 		printk(KERN_ERR
633 	"e820: cannot find a gap in the 32bit address range\n"
634 	"e820: PCI devices with unassigned 32bit BARs may break!\n");
635 #else
636 		gapstart = 0x10000000;
637 #endif
638 	}
639 
640 	/*
641 	 * e820_reserve_resources_late protect stolen RAM already
642 	 */
643 	pci_mem_start = gapstart;
644 
645 	printk(KERN_INFO
646 	       "e820: [mem %#010lx-%#010lx] available for PCI devices\n",
647 	       gapstart, gapstart + gapsize - 1);
648 }
649 
650 /*
651  * Called late during init, in free_initmem().
652  *
653  * Initial e820 and e820_saved are largish __initdata arrays.
654  * Copy them to (usually much smaller) dynamically allocated area.
655  * This is done after all tweaks we ever do to them:
656  * all functions which modify them are __init functions,
657  * they won't exist after this point.
658  */
659 __init void e820_reallocate_tables(void)
660 {
661 	struct e820map *n;
662 	int size;
663 
664 	size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820->nr_map;
665 	n = kmalloc(size, GFP_KERNEL);
666 	BUG_ON(!n);
667 	memcpy(n, e820, size);
668 	e820 = n;
669 
670 	size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820_saved->nr_map;
671 	n = kmalloc(size, GFP_KERNEL);
672 	BUG_ON(!n);
673 	memcpy(n, e820_saved, size);
674 	e820_saved = n;
675 }
676 
677 /**
678  * Because of the size limitation of struct boot_params, only first
679  * 128 E820 memory entries are passed to kernel via
680  * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
681  * linked list of struct setup_data, which is parsed here.
682  */
683 void __init parse_e820_ext(u64 phys_addr, u32 data_len)
684 {
685 	int entries;
686 	struct e820entry *extmap;
687 	struct setup_data *sdata;
688 
689 	sdata = early_memremap(phys_addr, data_len);
690 	entries = sdata->len / sizeof(struct e820entry);
691 	extmap = (struct e820entry *)(sdata->data);
692 	__append_e820_map(extmap, entries);
693 	sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map);
694 	early_memunmap(sdata, data_len);
695 	printk(KERN_INFO "e820: extended physical RAM map:\n");
696 	e820_print_map("extended");
697 }
698 
699 #if defined(CONFIG_X86_64) || \
700 	(defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
701 /**
702  * Find the ranges of physical addresses that do not correspond to
703  * e820 RAM areas and mark the corresponding pages as nosave for
704  * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
705  *
706  * This function requires the e820 map to be sorted and without any
707  * overlapping entries.
708  */
709 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
710 {
711 	int i;
712 	unsigned long pfn = 0;
713 
714 	for (i = 0; i < e820->nr_map; i++) {
715 		struct e820entry *ei = &e820->map[i];
716 
717 		if (pfn < PFN_UP(ei->addr))
718 			register_nosave_region(pfn, PFN_UP(ei->addr));
719 
720 		pfn = PFN_DOWN(ei->addr + ei->size);
721 
722 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
723 			register_nosave_region(PFN_UP(ei->addr), pfn);
724 
725 		if (pfn >= limit_pfn)
726 			break;
727 	}
728 }
729 #endif
730 
731 #ifdef CONFIG_ACPI
732 /**
733  * Mark ACPI NVS memory region, so that we can save/restore it during
734  * hibernation and the subsequent resume.
735  */
736 static int __init e820_mark_nvs_memory(void)
737 {
738 	int i;
739 
740 	for (i = 0; i < e820->nr_map; i++) {
741 		struct e820entry *ei = &e820->map[i];
742 
743 		if (ei->type == E820_NVS)
744 			acpi_nvs_register(ei->addr, ei->size);
745 	}
746 
747 	return 0;
748 }
749 core_initcall(e820_mark_nvs_memory);
750 #endif
751 
752 /*
753  * pre allocated 4k and reserved it in memblock and e820_saved
754  */
755 u64 __init early_reserve_e820(u64 size, u64 align)
756 {
757 	u64 addr;
758 
759 	addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
760 	if (addr) {
761 		e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED);
762 		printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n");
763 		update_e820_saved();
764 	}
765 
766 	return addr;
767 }
768 
769 #ifdef CONFIG_X86_32
770 # ifdef CONFIG_X86_PAE
771 #  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
772 # else
773 #  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
774 # endif
775 #else /* CONFIG_X86_32 */
776 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
777 #endif
778 
779 /*
780  * Find the highest page frame number we have available
781  */
782 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
783 {
784 	int i;
785 	unsigned long last_pfn = 0;
786 	unsigned long max_arch_pfn = MAX_ARCH_PFN;
787 
788 	for (i = 0; i < e820->nr_map; i++) {
789 		struct e820entry *ei = &e820->map[i];
790 		unsigned long start_pfn;
791 		unsigned long end_pfn;
792 
793 		if (ei->type != type)
794 			continue;
795 
796 		start_pfn = ei->addr >> PAGE_SHIFT;
797 		end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
798 
799 		if (start_pfn >= limit_pfn)
800 			continue;
801 		if (end_pfn > limit_pfn) {
802 			last_pfn = limit_pfn;
803 			break;
804 		}
805 		if (end_pfn > last_pfn)
806 			last_pfn = end_pfn;
807 	}
808 
809 	if (last_pfn > max_arch_pfn)
810 		last_pfn = max_arch_pfn;
811 
812 	printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
813 			 last_pfn, max_arch_pfn);
814 	return last_pfn;
815 }
816 unsigned long __init e820_end_of_ram_pfn(void)
817 {
818 	return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
819 }
820 
821 unsigned long __init e820_end_of_low_ram_pfn(void)
822 {
823 	return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_RAM);
824 }
825 
826 static void __init early_panic(char *msg)
827 {
828 	early_printk(msg);
829 	panic(msg);
830 }
831 
832 static int userdef __initdata;
833 
834 /* "mem=nopentium" disables the 4MB page tables. */
835 static int __init parse_memopt(char *p)
836 {
837 	u64 mem_size;
838 
839 	if (!p)
840 		return -EINVAL;
841 
842 	if (!strcmp(p, "nopentium")) {
843 #ifdef CONFIG_X86_32
844 		setup_clear_cpu_cap(X86_FEATURE_PSE);
845 		return 0;
846 #else
847 		printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
848 		return -EINVAL;
849 #endif
850 	}
851 
852 	userdef = 1;
853 	mem_size = memparse(p, &p);
854 	/* don't remove all of memory when handling "mem={invalid}" param */
855 	if (mem_size == 0)
856 		return -EINVAL;
857 	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
858 
859 	return 0;
860 }
861 early_param("mem", parse_memopt);
862 
863 static int __init parse_memmap_one(char *p)
864 {
865 	char *oldp;
866 	u64 start_at, mem_size;
867 
868 	if (!p)
869 		return -EINVAL;
870 
871 	if (!strncmp(p, "exactmap", 8)) {
872 #ifdef CONFIG_CRASH_DUMP
873 		/*
874 		 * If we are doing a crash dump, we still need to know
875 		 * the real mem size before original memory map is
876 		 * reset.
877 		 */
878 		saved_max_pfn = e820_end_of_ram_pfn();
879 #endif
880 		e820->nr_map = 0;
881 		userdef = 1;
882 		return 0;
883 	}
884 
885 	oldp = p;
886 	mem_size = memparse(p, &p);
887 	if (p == oldp)
888 		return -EINVAL;
889 
890 	userdef = 1;
891 	if (*p == '@') {
892 		start_at = memparse(p+1, &p);
893 		e820_add_region(start_at, mem_size, E820_RAM);
894 	} else if (*p == '#') {
895 		start_at = memparse(p+1, &p);
896 		e820_add_region(start_at, mem_size, E820_ACPI);
897 	} else if (*p == '$') {
898 		start_at = memparse(p+1, &p);
899 		e820_add_region(start_at, mem_size, E820_RESERVED);
900 	} else if (*p == '!') {
901 		start_at = memparse(p+1, &p);
902 		e820_add_region(start_at, mem_size, E820_PRAM);
903 	} else
904 		e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
905 
906 	return *p == '\0' ? 0 : -EINVAL;
907 }
908 static int __init parse_memmap_opt(char *str)
909 {
910 	while (str) {
911 		char *k = strchr(str, ',');
912 
913 		if (k)
914 			*k++ = 0;
915 
916 		parse_memmap_one(str);
917 		str = k;
918 	}
919 
920 	return 0;
921 }
922 early_param("memmap", parse_memmap_opt);
923 
924 void __init finish_e820_parsing(void)
925 {
926 	if (userdef) {
927 		if (sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map),
928 					&e820->nr_map) < 0)
929 			early_panic("Invalid user supplied memory map");
930 
931 		printk(KERN_INFO "e820: user-defined physical RAM map:\n");
932 		e820_print_map("user");
933 	}
934 }
935 
936 static const char *__init e820_type_to_string(int e820_type)
937 {
938 	switch (e820_type) {
939 	case E820_RESERVED_KERN:
940 	case E820_RAM:	return "System RAM";
941 	case E820_ACPI:	return "ACPI Tables";
942 	case E820_NVS:	return "ACPI Non-volatile Storage";
943 	case E820_UNUSABLE:	return "Unusable memory";
944 	case E820_PRAM: return "Persistent Memory (legacy)";
945 	case E820_PMEM: return "Persistent Memory";
946 	default:	return "reserved";
947 	}
948 }
949 
950 static unsigned long __init e820_type_to_iomem_type(int e820_type)
951 {
952 	switch (e820_type) {
953 	case E820_RESERVED_KERN:
954 	case E820_RAM:
955 		return IORESOURCE_SYSTEM_RAM;
956 	case E820_ACPI:
957 	case E820_NVS:
958 	case E820_UNUSABLE:
959 	case E820_PRAM:
960 	case E820_PMEM:
961 	default:
962 		return IORESOURCE_MEM;
963 	}
964 }
965 
966 static unsigned long __init e820_type_to_iores_desc(int e820_type)
967 {
968 	switch (e820_type) {
969 	case E820_ACPI:
970 		return IORES_DESC_ACPI_TABLES;
971 	case E820_NVS:
972 		return IORES_DESC_ACPI_NV_STORAGE;
973 	case E820_PMEM:
974 		return IORES_DESC_PERSISTENT_MEMORY;
975 	case E820_PRAM:
976 		return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
977 	case E820_RESERVED_KERN:
978 	case E820_RAM:
979 	case E820_UNUSABLE:
980 	default:
981 		return IORES_DESC_NONE;
982 	}
983 }
984 
985 static bool __init do_mark_busy(u32 type, struct resource *res)
986 {
987 	/* this is the legacy bios/dos rom-shadow + mmio region */
988 	if (res->start < (1ULL<<20))
989 		return true;
990 
991 	/*
992 	 * Treat persistent memory like device memory, i.e. reserve it
993 	 * for exclusive use of a driver
994 	 */
995 	switch (type) {
996 	case E820_RESERVED:
997 	case E820_PRAM:
998 	case E820_PMEM:
999 		return false;
1000 	default:
1001 		return true;
1002 	}
1003 }
1004 
1005 /*
1006  * Mark e820 reserved areas as busy for the resource manager.
1007  */
1008 static struct resource __initdata *e820_res;
1009 void __init e820_reserve_resources(void)
1010 {
1011 	int i;
1012 	struct resource *res;
1013 	u64 end;
1014 
1015 	res = alloc_bootmem(sizeof(struct resource) * e820->nr_map);
1016 	e820_res = res;
1017 	for (i = 0; i < e820->nr_map; i++) {
1018 		end = e820->map[i].addr + e820->map[i].size - 1;
1019 		if (end != (resource_size_t)end) {
1020 			res++;
1021 			continue;
1022 		}
1023 		res->name = e820_type_to_string(e820->map[i].type);
1024 		res->start = e820->map[i].addr;
1025 		res->end = end;
1026 
1027 		res->flags = e820_type_to_iomem_type(e820->map[i].type);
1028 		res->desc = e820_type_to_iores_desc(e820->map[i].type);
1029 
1030 		/*
1031 		 * don't register the region that could be conflicted with
1032 		 * pci device BAR resource and insert them later in
1033 		 * pcibios_resource_survey()
1034 		 */
1035 		if (do_mark_busy(e820->map[i].type, res)) {
1036 			res->flags |= IORESOURCE_BUSY;
1037 			insert_resource(&iomem_resource, res);
1038 		}
1039 		res++;
1040 	}
1041 
1042 	for (i = 0; i < e820_saved->nr_map; i++) {
1043 		struct e820entry *entry = &e820_saved->map[i];
1044 		firmware_map_add_early(entry->addr,
1045 			entry->addr + entry->size,
1046 			e820_type_to_string(entry->type));
1047 	}
1048 }
1049 
1050 /* How much should we pad RAM ending depending on where it is? */
1051 static unsigned long __init ram_alignment(resource_size_t pos)
1052 {
1053 	unsigned long mb = pos >> 20;
1054 
1055 	/* To 64kB in the first megabyte */
1056 	if (!mb)
1057 		return 64*1024;
1058 
1059 	/* To 1MB in the first 16MB */
1060 	if (mb < 16)
1061 		return 1024*1024;
1062 
1063 	/* To 64MB for anything above that */
1064 	return 64*1024*1024;
1065 }
1066 
1067 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1068 
1069 void __init e820_reserve_resources_late(void)
1070 {
1071 	int i;
1072 	struct resource *res;
1073 
1074 	res = e820_res;
1075 	for (i = 0; i < e820->nr_map; i++) {
1076 		if (!res->parent && res->end)
1077 			insert_resource_expand_to_fit(&iomem_resource, res);
1078 		res++;
1079 	}
1080 
1081 	/*
1082 	 * Try to bump up RAM regions to reasonable boundaries to
1083 	 * avoid stolen RAM:
1084 	 */
1085 	for (i = 0; i < e820->nr_map; i++) {
1086 		struct e820entry *entry = &e820->map[i];
1087 		u64 start, end;
1088 
1089 		if (entry->type != E820_RAM)
1090 			continue;
1091 		start = entry->addr + entry->size;
1092 		end = round_up(start, ram_alignment(start)) - 1;
1093 		if (end > MAX_RESOURCE_SIZE)
1094 			end = MAX_RESOURCE_SIZE;
1095 		if (start >= end)
1096 			continue;
1097 		printk(KERN_DEBUG
1098 		       "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n",
1099 		       start, end);
1100 		reserve_region_with_split(&iomem_resource, start, end,
1101 					  "RAM buffer");
1102 	}
1103 }
1104 
1105 char *__init default_machine_specific_memory_setup(void)
1106 {
1107 	char *who = "BIOS-e820";
1108 	u32 new_nr;
1109 	/*
1110 	 * Try to copy the BIOS-supplied E820-map.
1111 	 *
1112 	 * Otherwise fake a memory map; one section from 0k->640k,
1113 	 * the next section from 1mb->appropriate_mem_k
1114 	 */
1115 	new_nr = boot_params.e820_entries;
1116 	sanitize_e820_map(boot_params.e820_map,
1117 			ARRAY_SIZE(boot_params.e820_map),
1118 			&new_nr);
1119 	boot_params.e820_entries = new_nr;
1120 	if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1121 	  < 0) {
1122 		u64 mem_size;
1123 
1124 		/* compare results from other methods and take the greater */
1125 		if (boot_params.alt_mem_k
1126 		    < boot_params.screen_info.ext_mem_k) {
1127 			mem_size = boot_params.screen_info.ext_mem_k;
1128 			who = "BIOS-88";
1129 		} else {
1130 			mem_size = boot_params.alt_mem_k;
1131 			who = "BIOS-e801";
1132 		}
1133 
1134 		e820->nr_map = 0;
1135 		e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1136 		e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1137 	}
1138 
1139 	/* In case someone cares... */
1140 	return who;
1141 }
1142 
1143 void __init setup_memory_map(void)
1144 {
1145 	char *who;
1146 
1147 	who = x86_init.resources.memory_setup();
1148 	memcpy(e820_saved, e820, sizeof(struct e820map));
1149 	printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n");
1150 	e820_print_map(who);
1151 }
1152 
1153 void __init memblock_x86_fill(void)
1154 {
1155 	int i;
1156 	u64 end;
1157 
1158 	/*
1159 	 * EFI may have more than 128 entries
1160 	 * We are safe to enable resizing, beause memblock_x86_fill()
1161 	 * is rather later for x86
1162 	 */
1163 	memblock_allow_resize();
1164 
1165 	for (i = 0; i < e820->nr_map; i++) {
1166 		struct e820entry *ei = &e820->map[i];
1167 
1168 		end = ei->addr + ei->size;
1169 		if (end != (resource_size_t)end)
1170 			continue;
1171 
1172 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1173 			continue;
1174 
1175 		memblock_add(ei->addr, ei->size);
1176 	}
1177 
1178 	/* throw away partial pages */
1179 	memblock_trim_memory(PAGE_SIZE);
1180 
1181 	memblock_dump_all();
1182 }
1183 
1184 void __init memblock_find_dma_reserve(void)
1185 {
1186 #ifdef CONFIG_X86_64
1187 	u64 nr_pages = 0, nr_free_pages = 0;
1188 	unsigned long start_pfn, end_pfn;
1189 	phys_addr_t start, end;
1190 	int i;
1191 	u64 u;
1192 
1193 	/*
1194 	 * need to find out used area below MAX_DMA_PFN
1195 	 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1196 	 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1197 	 */
1198 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1199 		start_pfn = min(start_pfn, MAX_DMA_PFN);
1200 		end_pfn = min(end_pfn, MAX_DMA_PFN);
1201 		nr_pages += end_pfn - start_pfn;
1202 	}
1203 
1204 	for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1205 				NULL) {
1206 		start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1207 		end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1208 		if (start_pfn < end_pfn)
1209 			nr_free_pages += end_pfn - start_pfn;
1210 	}
1211 
1212 	set_dma_reserve(nr_pages - nr_free_pages);
1213 #endif
1214 }
1215