xref: /linux/arch/x86/kernel/e820.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
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 start_addr to end_addr.
584  */
585 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
586 		unsigned long start_addr, unsigned long long end_addr)
587 {
588 	unsigned long long last;
589 	int i = e820->nr_map;
590 	int found = 0;
591 
592 	last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
593 
594 	while (--i >= 0) {
595 		unsigned long long start = e820->map[i].addr;
596 		unsigned long long end = start + e820->map[i].size;
597 
598 		if (end < start_addr)
599 			continue;
600 
601 		/*
602 		 * Since "last" is at most 4GB, we know we'll
603 		 * fit in 32 bits if this condition is true
604 		 */
605 		if (last > end) {
606 			unsigned long gap = last - end;
607 
608 			if (gap >= *gapsize) {
609 				*gapsize = gap;
610 				*gapstart = end;
611 				found = 1;
612 			}
613 		}
614 		if (start < last)
615 			last = start;
616 	}
617 	return found;
618 }
619 
620 /*
621  * Search for the biggest gap in the low 32 bits of the e820
622  * memory space.  We pass this space to PCI to assign MMIO resources
623  * for hotplug or unconfigured devices in.
624  * Hopefully the BIOS let enough space left.
625  */
626 __init void e820_setup_gap(void)
627 {
628 	unsigned long gapstart, gapsize;
629 	int found;
630 
631 	gapstart = 0x10000000;
632 	gapsize = 0x400000;
633 	found  = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
634 
635 #ifdef CONFIG_X86_64
636 	if (!found) {
637 		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
638 		printk(KERN_ERR
639 	"e820: cannot find a gap in the 32bit address range\n"
640 	"e820: PCI devices with unassigned 32bit BARs may break!\n");
641 	}
642 #endif
643 
644 	/*
645 	 * e820_reserve_resources_late protect stolen RAM already
646 	 */
647 	pci_mem_start = gapstart;
648 
649 	printk(KERN_INFO
650 	       "e820: [mem %#010lx-%#010lx] available for PCI devices\n",
651 	       gapstart, gapstart + gapsize - 1);
652 }
653 
654 /*
655  * Called late during init, in free_initmem().
656  *
657  * Initial e820 and e820_saved are largish __initdata arrays.
658  * Copy them to (usually much smaller) dynamically allocated area.
659  * This is done after all tweaks we ever do to them:
660  * all functions which modify them are __init functions,
661  * they won't exist after this point.
662  */
663 __init void e820_reallocate_tables(void)
664 {
665 	struct e820map *n;
666 	int size;
667 
668 	size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820->nr_map;
669 	n = kmalloc(size, GFP_KERNEL);
670 	BUG_ON(!n);
671 	memcpy(n, e820, size);
672 	e820 = n;
673 
674 	size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820_saved->nr_map;
675 	n = kmalloc(size, GFP_KERNEL);
676 	BUG_ON(!n);
677 	memcpy(n, e820_saved, size);
678 	e820_saved = n;
679 }
680 
681 /**
682  * Because of the size limitation of struct boot_params, only first
683  * 128 E820 memory entries are passed to kernel via
684  * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
685  * linked list of struct setup_data, which is parsed here.
686  */
687 void __init parse_e820_ext(u64 phys_addr, u32 data_len)
688 {
689 	int entries;
690 	struct e820entry *extmap;
691 	struct setup_data *sdata;
692 
693 	sdata = early_memremap(phys_addr, data_len);
694 	entries = sdata->len / sizeof(struct e820entry);
695 	extmap = (struct e820entry *)(sdata->data);
696 	__append_e820_map(extmap, entries);
697 	sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map);
698 	early_memunmap(sdata, data_len);
699 	printk(KERN_INFO "e820: extended physical RAM map:\n");
700 	e820_print_map("extended");
701 }
702 
703 #if defined(CONFIG_X86_64) || \
704 	(defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
705 /**
706  * Find the ranges of physical addresses that do not correspond to
707  * e820 RAM areas and mark the corresponding pages as nosave for
708  * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
709  *
710  * This function requires the e820 map to be sorted and without any
711  * overlapping entries.
712  */
713 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
714 {
715 	int i;
716 	unsigned long pfn = 0;
717 
718 	for (i = 0; i < e820->nr_map; i++) {
719 		struct e820entry *ei = &e820->map[i];
720 
721 		if (pfn < PFN_UP(ei->addr))
722 			register_nosave_region(pfn, PFN_UP(ei->addr));
723 
724 		pfn = PFN_DOWN(ei->addr + ei->size);
725 
726 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
727 			register_nosave_region(PFN_UP(ei->addr), pfn);
728 
729 		if (pfn >= limit_pfn)
730 			break;
731 	}
732 }
733 #endif
734 
735 #ifdef CONFIG_ACPI
736 /**
737  * Mark ACPI NVS memory region, so that we can save/restore it during
738  * hibernation and the subsequent resume.
739  */
740 static int __init e820_mark_nvs_memory(void)
741 {
742 	int i;
743 
744 	for (i = 0; i < e820->nr_map; i++) {
745 		struct e820entry *ei = &e820->map[i];
746 
747 		if (ei->type == E820_NVS)
748 			acpi_nvs_register(ei->addr, ei->size);
749 	}
750 
751 	return 0;
752 }
753 core_initcall(e820_mark_nvs_memory);
754 #endif
755 
756 /*
757  * pre allocated 4k and reserved it in memblock and e820_saved
758  */
759 u64 __init early_reserve_e820(u64 size, u64 align)
760 {
761 	u64 addr;
762 
763 	addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
764 	if (addr) {
765 		e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED);
766 		printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n");
767 		update_e820_saved();
768 	}
769 
770 	return addr;
771 }
772 
773 #ifdef CONFIG_X86_32
774 # ifdef CONFIG_X86_PAE
775 #  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
776 # else
777 #  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
778 # endif
779 #else /* CONFIG_X86_32 */
780 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
781 #endif
782 
783 /*
784  * Find the highest page frame number we have available
785  */
786 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
787 {
788 	int i;
789 	unsigned long last_pfn = 0;
790 	unsigned long max_arch_pfn = MAX_ARCH_PFN;
791 
792 	for (i = 0; i < e820->nr_map; i++) {
793 		struct e820entry *ei = &e820->map[i];
794 		unsigned long start_pfn;
795 		unsigned long end_pfn;
796 
797 		if (ei->type != type)
798 			continue;
799 
800 		start_pfn = ei->addr >> PAGE_SHIFT;
801 		end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
802 
803 		if (start_pfn >= limit_pfn)
804 			continue;
805 		if (end_pfn > limit_pfn) {
806 			last_pfn = limit_pfn;
807 			break;
808 		}
809 		if (end_pfn > last_pfn)
810 			last_pfn = end_pfn;
811 	}
812 
813 	if (last_pfn > max_arch_pfn)
814 		last_pfn = max_arch_pfn;
815 
816 	printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
817 			 last_pfn, max_arch_pfn);
818 	return last_pfn;
819 }
820 unsigned long __init e820_end_of_ram_pfn(void)
821 {
822 	return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
823 }
824 
825 unsigned long __init e820_end_of_low_ram_pfn(void)
826 {
827 	return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_RAM);
828 }
829 
830 static void __init early_panic(char *msg)
831 {
832 	early_printk(msg);
833 	panic(msg);
834 }
835 
836 static int userdef __initdata;
837 
838 /* "mem=nopentium" disables the 4MB page tables. */
839 static int __init parse_memopt(char *p)
840 {
841 	u64 mem_size;
842 
843 	if (!p)
844 		return -EINVAL;
845 
846 	if (!strcmp(p, "nopentium")) {
847 #ifdef CONFIG_X86_32
848 		setup_clear_cpu_cap(X86_FEATURE_PSE);
849 		return 0;
850 #else
851 		printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
852 		return -EINVAL;
853 #endif
854 	}
855 
856 	userdef = 1;
857 	mem_size = memparse(p, &p);
858 	/* don't remove all of memory when handling "mem={invalid}" param */
859 	if (mem_size == 0)
860 		return -EINVAL;
861 	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
862 
863 	return 0;
864 }
865 early_param("mem", parse_memopt);
866 
867 static int __init parse_memmap_one(char *p)
868 {
869 	char *oldp;
870 	u64 start_at, mem_size;
871 
872 	if (!p)
873 		return -EINVAL;
874 
875 	if (!strncmp(p, "exactmap", 8)) {
876 #ifdef CONFIG_CRASH_DUMP
877 		/*
878 		 * If we are doing a crash dump, we still need to know
879 		 * the real mem size before original memory map is
880 		 * reset.
881 		 */
882 		saved_max_pfn = e820_end_of_ram_pfn();
883 #endif
884 		e820->nr_map = 0;
885 		userdef = 1;
886 		return 0;
887 	}
888 
889 	oldp = p;
890 	mem_size = memparse(p, &p);
891 	if (p == oldp)
892 		return -EINVAL;
893 
894 	userdef = 1;
895 	if (*p == '@') {
896 		start_at = memparse(p+1, &p);
897 		e820_add_region(start_at, mem_size, E820_RAM);
898 	} else if (*p == '#') {
899 		start_at = memparse(p+1, &p);
900 		e820_add_region(start_at, mem_size, E820_ACPI);
901 	} else if (*p == '$') {
902 		start_at = memparse(p+1, &p);
903 		e820_add_region(start_at, mem_size, E820_RESERVED);
904 	} else if (*p == '!') {
905 		start_at = memparse(p+1, &p);
906 		e820_add_region(start_at, mem_size, E820_PRAM);
907 	} else
908 		e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
909 
910 	return *p == '\0' ? 0 : -EINVAL;
911 }
912 static int __init parse_memmap_opt(char *str)
913 {
914 	while (str) {
915 		char *k = strchr(str, ',');
916 
917 		if (k)
918 			*k++ = 0;
919 
920 		parse_memmap_one(str);
921 		str = k;
922 	}
923 
924 	return 0;
925 }
926 early_param("memmap", parse_memmap_opt);
927 
928 void __init finish_e820_parsing(void)
929 {
930 	if (userdef) {
931 		if (sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map),
932 					&e820->nr_map) < 0)
933 			early_panic("Invalid user supplied memory map");
934 
935 		printk(KERN_INFO "e820: user-defined physical RAM map:\n");
936 		e820_print_map("user");
937 	}
938 }
939 
940 static const char *__init e820_type_to_string(int e820_type)
941 {
942 	switch (e820_type) {
943 	case E820_RESERVED_KERN:
944 	case E820_RAM:	return "System RAM";
945 	case E820_ACPI:	return "ACPI Tables";
946 	case E820_NVS:	return "ACPI Non-volatile Storage";
947 	case E820_UNUSABLE:	return "Unusable memory";
948 	case E820_PRAM: return "Persistent Memory (legacy)";
949 	case E820_PMEM: return "Persistent Memory";
950 	default:	return "reserved";
951 	}
952 }
953 
954 static unsigned long __init e820_type_to_iomem_type(int e820_type)
955 {
956 	switch (e820_type) {
957 	case E820_RESERVED_KERN:
958 	case E820_RAM:
959 		return IORESOURCE_SYSTEM_RAM;
960 	case E820_ACPI:
961 	case E820_NVS:
962 	case E820_UNUSABLE:
963 	case E820_PRAM:
964 	case E820_PMEM:
965 	default:
966 		return IORESOURCE_MEM;
967 	}
968 }
969 
970 static unsigned long __init e820_type_to_iores_desc(int e820_type)
971 {
972 	switch (e820_type) {
973 	case E820_ACPI:
974 		return IORES_DESC_ACPI_TABLES;
975 	case E820_NVS:
976 		return IORES_DESC_ACPI_NV_STORAGE;
977 	case E820_PMEM:
978 		return IORES_DESC_PERSISTENT_MEMORY;
979 	case E820_PRAM:
980 		return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
981 	case E820_RESERVED_KERN:
982 	case E820_RAM:
983 	case E820_UNUSABLE:
984 	default:
985 		return IORES_DESC_NONE;
986 	}
987 }
988 
989 static bool __init do_mark_busy(u32 type, struct resource *res)
990 {
991 	/* this is the legacy bios/dos rom-shadow + mmio region */
992 	if (res->start < (1ULL<<20))
993 		return true;
994 
995 	/*
996 	 * Treat persistent memory like device memory, i.e. reserve it
997 	 * for exclusive use of a driver
998 	 */
999 	switch (type) {
1000 	case E820_RESERVED:
1001 	case E820_PRAM:
1002 	case E820_PMEM:
1003 		return false;
1004 	default:
1005 		return true;
1006 	}
1007 }
1008 
1009 /*
1010  * Mark e820 reserved areas as busy for the resource manager.
1011  */
1012 static struct resource __initdata *e820_res;
1013 void __init e820_reserve_resources(void)
1014 {
1015 	int i;
1016 	struct resource *res;
1017 	u64 end;
1018 
1019 	res = alloc_bootmem(sizeof(struct resource) * e820->nr_map);
1020 	e820_res = res;
1021 	for (i = 0; i < e820->nr_map; i++) {
1022 		end = e820->map[i].addr + e820->map[i].size - 1;
1023 		if (end != (resource_size_t)end) {
1024 			res++;
1025 			continue;
1026 		}
1027 		res->name = e820_type_to_string(e820->map[i].type);
1028 		res->start = e820->map[i].addr;
1029 		res->end = end;
1030 
1031 		res->flags = e820_type_to_iomem_type(e820->map[i].type);
1032 		res->desc = e820_type_to_iores_desc(e820->map[i].type);
1033 
1034 		/*
1035 		 * don't register the region that could be conflicted with
1036 		 * pci device BAR resource and insert them later in
1037 		 * pcibios_resource_survey()
1038 		 */
1039 		if (do_mark_busy(e820->map[i].type, res)) {
1040 			res->flags |= IORESOURCE_BUSY;
1041 			insert_resource(&iomem_resource, res);
1042 		}
1043 		res++;
1044 	}
1045 
1046 	for (i = 0; i < e820_saved->nr_map; i++) {
1047 		struct e820entry *entry = &e820_saved->map[i];
1048 		firmware_map_add_early(entry->addr,
1049 			entry->addr + entry->size,
1050 			e820_type_to_string(entry->type));
1051 	}
1052 }
1053 
1054 /* How much should we pad RAM ending depending on where it is? */
1055 static unsigned long __init ram_alignment(resource_size_t pos)
1056 {
1057 	unsigned long mb = pos >> 20;
1058 
1059 	/* To 64kB in the first megabyte */
1060 	if (!mb)
1061 		return 64*1024;
1062 
1063 	/* To 1MB in the first 16MB */
1064 	if (mb < 16)
1065 		return 1024*1024;
1066 
1067 	/* To 64MB for anything above that */
1068 	return 64*1024*1024;
1069 }
1070 
1071 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1072 
1073 void __init e820_reserve_resources_late(void)
1074 {
1075 	int i;
1076 	struct resource *res;
1077 
1078 	res = e820_res;
1079 	for (i = 0; i < e820->nr_map; i++) {
1080 		if (!res->parent && res->end)
1081 			insert_resource_expand_to_fit(&iomem_resource, res);
1082 		res++;
1083 	}
1084 
1085 	/*
1086 	 * Try to bump up RAM regions to reasonable boundaries to
1087 	 * avoid stolen RAM:
1088 	 */
1089 	for (i = 0; i < e820->nr_map; i++) {
1090 		struct e820entry *entry = &e820->map[i];
1091 		u64 start, end;
1092 
1093 		if (entry->type != E820_RAM)
1094 			continue;
1095 		start = entry->addr + entry->size;
1096 		end = round_up(start, ram_alignment(start)) - 1;
1097 		if (end > MAX_RESOURCE_SIZE)
1098 			end = MAX_RESOURCE_SIZE;
1099 		if (start >= end)
1100 			continue;
1101 		printk(KERN_DEBUG
1102 		       "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n",
1103 		       start, end);
1104 		reserve_region_with_split(&iomem_resource, start, end,
1105 					  "RAM buffer");
1106 	}
1107 }
1108 
1109 char *__init default_machine_specific_memory_setup(void)
1110 {
1111 	char *who = "BIOS-e820";
1112 	u32 new_nr;
1113 	/*
1114 	 * Try to copy the BIOS-supplied E820-map.
1115 	 *
1116 	 * Otherwise fake a memory map; one section from 0k->640k,
1117 	 * the next section from 1mb->appropriate_mem_k
1118 	 */
1119 	new_nr = boot_params.e820_entries;
1120 	sanitize_e820_map(boot_params.e820_map,
1121 			ARRAY_SIZE(boot_params.e820_map),
1122 			&new_nr);
1123 	boot_params.e820_entries = new_nr;
1124 	if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1125 	  < 0) {
1126 		u64 mem_size;
1127 
1128 		/* compare results from other methods and take the greater */
1129 		if (boot_params.alt_mem_k
1130 		    < boot_params.screen_info.ext_mem_k) {
1131 			mem_size = boot_params.screen_info.ext_mem_k;
1132 			who = "BIOS-88";
1133 		} else {
1134 			mem_size = boot_params.alt_mem_k;
1135 			who = "BIOS-e801";
1136 		}
1137 
1138 		e820->nr_map = 0;
1139 		e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1140 		e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1141 	}
1142 
1143 	/* In case someone cares... */
1144 	return who;
1145 }
1146 
1147 void __init setup_memory_map(void)
1148 {
1149 	char *who;
1150 
1151 	who = x86_init.resources.memory_setup();
1152 	memcpy(e820_saved, e820, sizeof(struct e820map));
1153 	printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n");
1154 	e820_print_map(who);
1155 }
1156 
1157 void __init memblock_x86_fill(void)
1158 {
1159 	int i;
1160 	u64 end;
1161 
1162 	/*
1163 	 * EFI may have more than 128 entries
1164 	 * We are safe to enable resizing, beause memblock_x86_fill()
1165 	 * is rather later for x86
1166 	 */
1167 	memblock_allow_resize();
1168 
1169 	for (i = 0; i < e820->nr_map; i++) {
1170 		struct e820entry *ei = &e820->map[i];
1171 
1172 		end = ei->addr + ei->size;
1173 		if (end != (resource_size_t)end)
1174 			continue;
1175 
1176 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1177 			continue;
1178 
1179 		memblock_add(ei->addr, ei->size);
1180 	}
1181 
1182 	/* throw away partial pages */
1183 	memblock_trim_memory(PAGE_SIZE);
1184 
1185 	memblock_dump_all();
1186 }
1187 
1188 void __init memblock_find_dma_reserve(void)
1189 {
1190 #ifdef CONFIG_X86_64
1191 	u64 nr_pages = 0, nr_free_pages = 0;
1192 	unsigned long start_pfn, end_pfn;
1193 	phys_addr_t start, end;
1194 	int i;
1195 	u64 u;
1196 
1197 	/*
1198 	 * need to find out used area below MAX_DMA_PFN
1199 	 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1200 	 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1201 	 */
1202 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1203 		start_pfn = min(start_pfn, MAX_DMA_PFN);
1204 		end_pfn = min(end_pfn, MAX_DMA_PFN);
1205 		nr_pages += end_pfn - start_pfn;
1206 	}
1207 
1208 	for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1209 				NULL) {
1210 		start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1211 		end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1212 		if (start_pfn < end_pfn)
1213 			nr_free_pages += end_pfn - start_pfn;
1214 	}
1215 
1216 	set_dma_reserve(nr_pages - nr_free_pages);
1217 #endif
1218 }
1219