xref: /openbmc/linux/arch/x86/kernel/e820.c (revision baa7eb025ab14f3cba2e35c0a8648f9c9f01d24f)
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/bootmem.h>
15 #include <linux/pfn.h>
16 #include <linux/suspend.h>
17 #include <linux/firmware-map.h>
18 #include <linux/memblock.h>
19 
20 #include <asm/e820.h>
21 #include <asm/proto.h>
22 #include <asm/setup.h>
23 
24 /*
25  * The e820 map is the map that gets modified e.g. with command line parameters
26  * and that is also registered with modifications in the kernel resource tree
27  * with the iomem_resource as parent.
28  *
29  * The e820_saved is directly saved after the BIOS-provided memory map is
30  * copied. It doesn't get modified afterwards. It's registered for the
31  * /sys/firmware/memmap interface.
32  *
33  * That memory map is not modified and is used as base for kexec. The kexec'd
34  * kernel should get the same memory map as the firmware provides. Then the
35  * user can e.g. boot the original kernel with mem=1G while still booting the
36  * next kernel with full memory.
37  */
38 struct e820map e820;
39 struct e820map e820_saved;
40 
41 /* For PCI or other memory-mapped resources */
42 unsigned long pci_mem_start = 0xaeedbabe;
43 #ifdef CONFIG_PCI
44 EXPORT_SYMBOL(pci_mem_start);
45 #endif
46 
47 /*
48  * This function checks if any part of the range <start,end> is mapped
49  * with type.
50  */
51 int
52 e820_any_mapped(u64 start, u64 end, unsigned type)
53 {
54 	int i;
55 
56 	for (i = 0; i < e820.nr_map; i++) {
57 		struct e820entry *ei = &e820.map[i];
58 
59 		if (type && ei->type != type)
60 			continue;
61 		if (ei->addr >= end || ei->addr + ei->size <= start)
62 			continue;
63 		return 1;
64 	}
65 	return 0;
66 }
67 EXPORT_SYMBOL_GPL(e820_any_mapped);
68 
69 /*
70  * This function checks if the entire range <start,end> is mapped with type.
71  *
72  * Note: this function only works correct if the e820 table is sorted and
73  * not-overlapping, which is the case
74  */
75 int __init e820_all_mapped(u64 start, u64 end, unsigned type)
76 {
77 	int i;
78 
79 	for (i = 0; i < e820.nr_map; i++) {
80 		struct e820entry *ei = &e820.map[i];
81 
82 		if (type && ei->type != type)
83 			continue;
84 		/* is the region (part) in overlap with the current region ?*/
85 		if (ei->addr >= end || ei->addr + ei->size <= start)
86 			continue;
87 
88 		/* if the region is at the beginning of <start,end> we move
89 		 * start to the end of the region since it's ok until there
90 		 */
91 		if (ei->addr <= start)
92 			start = ei->addr + ei->size;
93 		/*
94 		 * if start is now at or beyond end, we're done, full
95 		 * coverage
96 		 */
97 		if (start >= end)
98 			return 1;
99 	}
100 	return 0;
101 }
102 
103 /*
104  * Add a memory region to the kernel e820 map.
105  */
106 static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
107 					 int type)
108 {
109 	int x = e820x->nr_map;
110 
111 	if (x >= ARRAY_SIZE(e820x->map)) {
112 		printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
113 		return;
114 	}
115 
116 	e820x->map[x].addr = start;
117 	e820x->map[x].size = size;
118 	e820x->map[x].type = type;
119 	e820x->nr_map++;
120 }
121 
122 void __init e820_add_region(u64 start, u64 size, int type)
123 {
124 	__e820_add_region(&e820, start, size, type);
125 }
126 
127 static void __init e820_print_type(u32 type)
128 {
129 	switch (type) {
130 	case E820_RAM:
131 	case E820_RESERVED_KERN:
132 		printk(KERN_CONT "(usable)");
133 		break;
134 	case E820_RESERVED:
135 		printk(KERN_CONT "(reserved)");
136 		break;
137 	case E820_ACPI:
138 		printk(KERN_CONT "(ACPI data)");
139 		break;
140 	case E820_NVS:
141 		printk(KERN_CONT "(ACPI NVS)");
142 		break;
143 	case E820_UNUSABLE:
144 		printk(KERN_CONT "(unusable)");
145 		break;
146 	default:
147 		printk(KERN_CONT "type %u", type);
148 		break;
149 	}
150 }
151 
152 void __init e820_print_map(char *who)
153 {
154 	int i;
155 
156 	for (i = 0; i < e820.nr_map; i++) {
157 		printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
158 		       (unsigned long long) e820.map[i].addr,
159 		       (unsigned long long)
160 		       (e820.map[i].addr + e820.map[i].size));
161 		e820_print_type(e820.map[i].type);
162 		printk(KERN_CONT "\n");
163 	}
164 }
165 
166 /*
167  * Sanitize the BIOS e820 map.
168  *
169  * Some e820 responses include overlapping entries. The following
170  * replaces the original e820 map with a new one, removing overlaps,
171  * and resolving conflicting memory types in favor of highest
172  * numbered type.
173  *
174  * The input parameter biosmap points to an array of 'struct
175  * e820entry' which on entry has elements in the range [0, *pnr_map)
176  * valid, and which has space for up to max_nr_map entries.
177  * On return, the resulting sanitized e820 map entries will be in
178  * overwritten in the same location, starting at biosmap.
179  *
180  * The integer pointed to by pnr_map must be valid on entry (the
181  * current number of valid entries located at biosmap) and will
182  * be updated on return, with the new number of valid entries
183  * (something no more than max_nr_map.)
184  *
185  * The return value from sanitize_e820_map() is zero if it
186  * successfully 'sanitized' the map entries passed in, and is -1
187  * if it did nothing, which can happen if either of (1) it was
188  * only passed one map entry, or (2) any of the input map entries
189  * were invalid (start + size < start, meaning that the size was
190  * so big the described memory range wrapped around through zero.)
191  *
192  *	Visually we're performing the following
193  *	(1,2,3,4 = memory types)...
194  *
195  *	Sample memory map (w/overlaps):
196  *	   ____22__________________
197  *	   ______________________4_
198  *	   ____1111________________
199  *	   _44_____________________
200  *	   11111111________________
201  *	   ____________________33__
202  *	   ___________44___________
203  *	   __________33333_________
204  *	   ______________22________
205  *	   ___________________2222_
206  *	   _________111111111______
207  *	   _____________________11_
208  *	   _________________4______
209  *
210  *	Sanitized equivalent (no overlap):
211  *	   1_______________________
212  *	   _44_____________________
213  *	   ___1____________________
214  *	   ____22__________________
215  *	   ______11________________
216  *	   _________1______________
217  *	   __________3_____________
218  *	   ___________44___________
219  *	   _____________33_________
220  *	   _______________2________
221  *	   ________________1_______
222  *	   _________________4______
223  *	   ___________________2____
224  *	   ____________________33__
225  *	   ______________________4_
226  */
227 
228 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
229 			     u32 *pnr_map)
230 {
231 	struct change_member {
232 		struct e820entry *pbios; /* pointer to original bios entry */
233 		unsigned long long addr; /* address for this change point */
234 	};
235 	static struct change_member change_point_list[2*E820_X_MAX] __initdata;
236 	static struct change_member *change_point[2*E820_X_MAX] __initdata;
237 	static struct e820entry *overlap_list[E820_X_MAX] __initdata;
238 	static struct e820entry new_bios[E820_X_MAX] __initdata;
239 	struct change_member *change_tmp;
240 	unsigned long current_type, last_type;
241 	unsigned long long last_addr;
242 	int chgidx, still_changing;
243 	int overlap_entries;
244 	int new_bios_entry;
245 	int old_nr, new_nr, chg_nr;
246 	int i;
247 
248 	/* if there's only one memory region, don't bother */
249 	if (*pnr_map < 2)
250 		return -1;
251 
252 	old_nr = *pnr_map;
253 	BUG_ON(old_nr > max_nr_map);
254 
255 	/* bail out if we find any unreasonable addresses in bios map */
256 	for (i = 0; i < old_nr; i++)
257 		if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
258 			return -1;
259 
260 	/* create pointers for initial change-point information (for sorting) */
261 	for (i = 0; i < 2 * old_nr; i++)
262 		change_point[i] = &change_point_list[i];
263 
264 	/* record all known change-points (starting and ending addresses),
265 	   omitting those that are for empty memory regions */
266 	chgidx = 0;
267 	for (i = 0; i < old_nr; i++)	{
268 		if (biosmap[i].size != 0) {
269 			change_point[chgidx]->addr = biosmap[i].addr;
270 			change_point[chgidx++]->pbios = &biosmap[i];
271 			change_point[chgidx]->addr = biosmap[i].addr +
272 				biosmap[i].size;
273 			change_point[chgidx++]->pbios = &biosmap[i];
274 		}
275 	}
276 	chg_nr = chgidx;
277 
278 	/* sort change-point list by memory addresses (low -> high) */
279 	still_changing = 1;
280 	while (still_changing)	{
281 		still_changing = 0;
282 		for (i = 1; i < chg_nr; i++)  {
283 			unsigned long long curaddr, lastaddr;
284 			unsigned long long curpbaddr, lastpbaddr;
285 
286 			curaddr = change_point[i]->addr;
287 			lastaddr = change_point[i - 1]->addr;
288 			curpbaddr = change_point[i]->pbios->addr;
289 			lastpbaddr = change_point[i - 1]->pbios->addr;
290 
291 			/*
292 			 * swap entries, when:
293 			 *
294 			 * curaddr > lastaddr or
295 			 * curaddr == lastaddr and curaddr == curpbaddr and
296 			 * lastaddr != lastpbaddr
297 			 */
298 			if (curaddr < lastaddr ||
299 			    (curaddr == lastaddr && curaddr == curpbaddr &&
300 			     lastaddr != lastpbaddr)) {
301 				change_tmp = change_point[i];
302 				change_point[i] = change_point[i-1];
303 				change_point[i-1] = change_tmp;
304 				still_changing = 1;
305 			}
306 		}
307 	}
308 
309 	/* create a new bios memory map, removing overlaps */
310 	overlap_entries = 0;	 /* number of entries in the overlap table */
311 	new_bios_entry = 0;	 /* index for creating new bios map entries */
312 	last_type = 0;		 /* start with undefined memory type */
313 	last_addr = 0;		 /* start with 0 as last starting address */
314 
315 	/* loop through change-points, determining affect on the new bios map */
316 	for (chgidx = 0; chgidx < chg_nr; chgidx++) {
317 		/* keep track of all overlapping bios entries */
318 		if (change_point[chgidx]->addr ==
319 		    change_point[chgidx]->pbios->addr) {
320 			/*
321 			 * add map entry to overlap list (> 1 entry
322 			 * implies an overlap)
323 			 */
324 			overlap_list[overlap_entries++] =
325 				change_point[chgidx]->pbios;
326 		} else {
327 			/*
328 			 * remove entry from list (order independent,
329 			 * so swap with last)
330 			 */
331 			for (i = 0; i < overlap_entries; i++) {
332 				if (overlap_list[i] ==
333 				    change_point[chgidx]->pbios)
334 					overlap_list[i] =
335 						overlap_list[overlap_entries-1];
336 			}
337 			overlap_entries--;
338 		}
339 		/*
340 		 * if there are overlapping entries, decide which
341 		 * "type" to use (larger value takes precedence --
342 		 * 1=usable, 2,3,4,4+=unusable)
343 		 */
344 		current_type = 0;
345 		for (i = 0; i < overlap_entries; i++)
346 			if (overlap_list[i]->type > current_type)
347 				current_type = overlap_list[i]->type;
348 		/*
349 		 * continue building up new bios map based on this
350 		 * information
351 		 */
352 		if (current_type != last_type)	{
353 			if (last_type != 0)	 {
354 				new_bios[new_bios_entry].size =
355 					change_point[chgidx]->addr - last_addr;
356 				/*
357 				 * move forward only if the new size
358 				 * was non-zero
359 				 */
360 				if (new_bios[new_bios_entry].size != 0)
361 					/*
362 					 * no more space left for new
363 					 * bios entries ?
364 					 */
365 					if (++new_bios_entry >= max_nr_map)
366 						break;
367 			}
368 			if (current_type != 0)	{
369 				new_bios[new_bios_entry].addr =
370 					change_point[chgidx]->addr;
371 				new_bios[new_bios_entry].type = current_type;
372 				last_addr = change_point[chgidx]->addr;
373 			}
374 			last_type = current_type;
375 		}
376 	}
377 	/* retain count for new bios entries */
378 	new_nr = new_bios_entry;
379 
380 	/* copy new bios mapping into original location */
381 	memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
382 	*pnr_map = new_nr;
383 
384 	return 0;
385 }
386 
387 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
388 {
389 	while (nr_map) {
390 		u64 start = biosmap->addr;
391 		u64 size = biosmap->size;
392 		u64 end = start + size;
393 		u32 type = biosmap->type;
394 
395 		/* Overflow in 64 bits? Ignore the memory map. */
396 		if (start > end)
397 			return -1;
398 
399 		e820_add_region(start, size, type);
400 
401 		biosmap++;
402 		nr_map--;
403 	}
404 	return 0;
405 }
406 
407 /*
408  * Copy the BIOS e820 map into a safe place.
409  *
410  * Sanity-check it while we're at it..
411  *
412  * If we're lucky and live on a modern system, the setup code
413  * will have given us a memory map that we can use to properly
414  * set up memory.  If we aren't, we'll fake a memory map.
415  */
416 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
417 {
418 	/* Only one memory region (or negative)? Ignore it */
419 	if (nr_map < 2)
420 		return -1;
421 
422 	return __append_e820_map(biosmap, nr_map);
423 }
424 
425 static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
426 					u64 size, unsigned old_type,
427 					unsigned new_type)
428 {
429 	u64 end;
430 	unsigned int i;
431 	u64 real_updated_size = 0;
432 
433 	BUG_ON(old_type == new_type);
434 
435 	if (size > (ULLONG_MAX - start))
436 		size = ULLONG_MAX - start;
437 
438 	end = start + size;
439 	printk(KERN_DEBUG "e820 update range: %016Lx - %016Lx ",
440 		       (unsigned long long) start,
441 		       (unsigned long long) end);
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 range: %016Lx - %016Lx ",
521 		       (unsigned long long) start,
522 		       (unsigned long long) end);
523 	if (checktype)
524 		e820_print_type(old_type);
525 	printk(KERN_CONT "\n");
526 
527 	for (i = 0; i < e820.nr_map; i++) {
528 		struct e820entry *ei = &e820.map[i];
529 		u64 final_start, final_end;
530 		u64 ei_end;
531 
532 		if (checktype && ei->type != old_type)
533 			continue;
534 
535 		ei_end = ei->addr + ei->size;
536 		/* totally covered? */
537 		if (ei->addr >= start && ei_end <= end) {
538 			real_removed_size += ei->size;
539 			memset(ei, 0, sizeof(struct e820entry));
540 			continue;
541 		}
542 
543 		/* new range is totally covered? */
544 		if (ei->addr < start && ei_end > end) {
545 			e820_add_region(end, ei_end - end, ei->type);
546 			ei->size = start - ei->addr;
547 			real_removed_size += size;
548 			continue;
549 		}
550 
551 		/* partially covered */
552 		final_start = max(start, ei->addr);
553 		final_end = min(end, ei_end);
554 		if (final_start >= final_end)
555 			continue;
556 		real_removed_size += final_end - final_start;
557 
558 		/*
559 		 * left range could be head or tail, so need to update
560 		 * size at first.
561 		 */
562 		ei->size -= final_end - final_start;
563 		if (ei->addr < final_start)
564 			continue;
565 		ei->addr = final_end;
566 	}
567 	return real_removed_size;
568 }
569 
570 void __init update_e820(void)
571 {
572 	u32 nr_map;
573 
574 	nr_map = e820.nr_map;
575 	if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
576 		return;
577 	e820.nr_map = nr_map;
578 	printk(KERN_INFO "modified physical RAM map:\n");
579 	e820_print_map("modified");
580 }
581 static void __init update_e820_saved(void)
582 {
583 	u32 nr_map;
584 
585 	nr_map = e820_saved.nr_map;
586 	if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
587 		return;
588 	e820_saved.nr_map = nr_map;
589 }
590 #define MAX_GAP_END 0x100000000ull
591 /*
592  * Search for a gap in the e820 memory space from start_addr to end_addr.
593  */
594 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
595 		unsigned long start_addr, unsigned long long end_addr)
596 {
597 	unsigned long long last;
598 	int i = e820.nr_map;
599 	int found = 0;
600 
601 	last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
602 
603 	while (--i >= 0) {
604 		unsigned long long start = e820.map[i].addr;
605 		unsigned long long end = start + e820.map[i].size;
606 
607 		if (end < start_addr)
608 			continue;
609 
610 		/*
611 		 * Since "last" is at most 4GB, we know we'll
612 		 * fit in 32 bits if this condition is true
613 		 */
614 		if (last > end) {
615 			unsigned long gap = last - end;
616 
617 			if (gap >= *gapsize) {
618 				*gapsize = gap;
619 				*gapstart = end;
620 				found = 1;
621 			}
622 		}
623 		if (start < last)
624 			last = start;
625 	}
626 	return found;
627 }
628 
629 /*
630  * Search for the biggest gap in the low 32 bits of the e820
631  * memory space.  We pass this space to PCI to assign MMIO resources
632  * for hotplug or unconfigured devices in.
633  * Hopefully the BIOS let enough space left.
634  */
635 __init void e820_setup_gap(void)
636 {
637 	unsigned long gapstart, gapsize;
638 	int found;
639 
640 	gapstart = 0x10000000;
641 	gapsize = 0x400000;
642 	found  = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
643 
644 #ifdef CONFIG_X86_64
645 	if (!found) {
646 		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
647 		printk(KERN_ERR
648 	"PCI: Warning: Cannot find a gap in the 32bit address range\n"
649 	"PCI: Unassigned devices with 32bit resource registers may break!\n");
650 	}
651 #endif
652 
653 	/*
654 	 * e820_reserve_resources_late protect stolen RAM already
655 	 */
656 	pci_mem_start = gapstart;
657 
658 	printk(KERN_INFO
659 	       "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
660 	       pci_mem_start, gapstart, gapsize);
661 }
662 
663 /**
664  * Because of the size limitation of struct boot_params, only first
665  * 128 E820 memory entries are passed to kernel via
666  * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
667  * linked list of struct setup_data, which is parsed here.
668  */
669 void __init parse_e820_ext(struct setup_data *sdata, unsigned long pa_data)
670 {
671 	u32 map_len;
672 	int entries;
673 	struct e820entry *extmap;
674 
675 	entries = sdata->len / sizeof(struct e820entry);
676 	map_len = sdata->len + sizeof(struct setup_data);
677 	if (map_len > PAGE_SIZE)
678 		sdata = early_ioremap(pa_data, map_len);
679 	extmap = (struct e820entry *)(sdata->data);
680 	__append_e820_map(extmap, entries);
681 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
682 	if (map_len > PAGE_SIZE)
683 		early_iounmap(sdata, map_len);
684 	printk(KERN_INFO "extended physical RAM map:\n");
685 	e820_print_map("extended");
686 }
687 
688 #if defined(CONFIG_X86_64) || \
689 	(defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
690 /**
691  * Find the ranges of physical addresses that do not correspond to
692  * e820 RAM areas and mark the corresponding pages as nosave for
693  * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
694  *
695  * This function requires the e820 map to be sorted and without any
696  * overlapping entries and assumes the first e820 area to be RAM.
697  */
698 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
699 {
700 	int i;
701 	unsigned long pfn;
702 
703 	pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
704 	for (i = 1; i < e820.nr_map; i++) {
705 		struct e820entry *ei = &e820.map[i];
706 
707 		if (pfn < PFN_UP(ei->addr))
708 			register_nosave_region(pfn, PFN_UP(ei->addr));
709 
710 		pfn = PFN_DOWN(ei->addr + ei->size);
711 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
712 			register_nosave_region(PFN_UP(ei->addr), pfn);
713 
714 		if (pfn >= limit_pfn)
715 			break;
716 	}
717 }
718 #endif
719 
720 #ifdef CONFIG_HIBERNATION
721 /**
722  * Mark ACPI NVS memory region, so that we can save/restore it during
723  * hibernation and the subsequent resume.
724  */
725 static int __init e820_mark_nvs_memory(void)
726 {
727 	int i;
728 
729 	for (i = 0; i < e820.nr_map; i++) {
730 		struct e820entry *ei = &e820.map[i];
731 
732 		if (ei->type == E820_NVS)
733 			suspend_nvs_register(ei->addr, ei->size);
734 	}
735 
736 	return 0;
737 }
738 core_initcall(e820_mark_nvs_memory);
739 #endif
740 
741 /*
742  * pre allocated 4k and reserved it in memblock and e820_saved
743  */
744 u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
745 {
746 	u64 size = 0;
747 	u64 addr;
748 	u64 start;
749 
750 	for (start = startt; ; start += size) {
751 		start = memblock_x86_find_in_range_size(start, &size, align);
752 		if (start == MEMBLOCK_ERROR)
753 			return 0;
754 		if (size >= sizet)
755 			break;
756 	}
757 
758 #ifdef CONFIG_X86_32
759 	if (start >= MAXMEM)
760 		return 0;
761 	if (start + size > MAXMEM)
762 		size = MAXMEM - start;
763 #endif
764 
765 	addr = round_down(start + size - sizet, align);
766 	if (addr < start)
767 		return 0;
768 	memblock_x86_reserve_range(addr, addr + sizet, "new next");
769 	e820_update_range_saved(addr, sizet, E820_RAM, E820_RESERVED);
770 	printk(KERN_INFO "update e820_saved for early_reserve_e820\n");
771 	update_e820_saved();
772 
773 	return addr;
774 }
775 
776 #ifdef CONFIG_X86_32
777 # ifdef CONFIG_X86_PAE
778 #  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
779 # else
780 #  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
781 # endif
782 #else /* CONFIG_X86_32 */
783 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
784 #endif
785 
786 /*
787  * Find the highest page frame number we have available
788  */
789 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
790 {
791 	int i;
792 	unsigned long last_pfn = 0;
793 	unsigned long max_arch_pfn = MAX_ARCH_PFN;
794 
795 	for (i = 0; i < e820.nr_map; i++) {
796 		struct e820entry *ei = &e820.map[i];
797 		unsigned long start_pfn;
798 		unsigned long end_pfn;
799 
800 		if (ei->type != type)
801 			continue;
802 
803 		start_pfn = ei->addr >> PAGE_SHIFT;
804 		end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
805 
806 		if (start_pfn >= limit_pfn)
807 			continue;
808 		if (end_pfn > limit_pfn) {
809 			last_pfn = limit_pfn;
810 			break;
811 		}
812 		if (end_pfn > last_pfn)
813 			last_pfn = end_pfn;
814 	}
815 
816 	if (last_pfn > max_arch_pfn)
817 		last_pfn = max_arch_pfn;
818 
819 	printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
820 			 last_pfn, max_arch_pfn);
821 	return last_pfn;
822 }
823 unsigned long __init e820_end_of_ram_pfn(void)
824 {
825 	return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
826 }
827 
828 unsigned long __init e820_end_of_low_ram_pfn(void)
829 {
830 	return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
831 }
832 
833 static void early_panic(char *msg)
834 {
835 	early_printk(msg);
836 	panic(msg);
837 }
838 
839 static int userdef __initdata;
840 
841 /* "mem=nopentium" disables the 4MB page tables. */
842 static int __init parse_memopt(char *p)
843 {
844 	u64 mem_size;
845 
846 	if (!p)
847 		return -EINVAL;
848 
849 #ifdef CONFIG_X86_32
850 	if (!strcmp(p, "nopentium")) {
851 		setup_clear_cpu_cap(X86_FEATURE_PSE);
852 		return 0;
853 	}
854 #endif
855 
856 	userdef = 1;
857 	mem_size = memparse(p, &p);
858 	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
859 
860 	return 0;
861 }
862 early_param("mem", parse_memopt);
863 
864 static int __init parse_memmap_opt(char *p)
865 {
866 	char *oldp;
867 	u64 start_at, mem_size;
868 
869 	if (!p)
870 		return -EINVAL;
871 
872 	if (!strncmp(p, "exactmap", 8)) {
873 #ifdef CONFIG_CRASH_DUMP
874 		/*
875 		 * If we are doing a crash dump, we still need to know
876 		 * the real mem size before original memory map is
877 		 * reset.
878 		 */
879 		saved_max_pfn = e820_end_of_ram_pfn();
880 #endif
881 		e820.nr_map = 0;
882 		userdef = 1;
883 		return 0;
884 	}
885 
886 	oldp = p;
887 	mem_size = memparse(p, &p);
888 	if (p == oldp)
889 		return -EINVAL;
890 
891 	userdef = 1;
892 	if (*p == '@') {
893 		start_at = memparse(p+1, &p);
894 		e820_add_region(start_at, mem_size, E820_RAM);
895 	} else if (*p == '#') {
896 		start_at = memparse(p+1, &p);
897 		e820_add_region(start_at, mem_size, E820_ACPI);
898 	} else if (*p == '$') {
899 		start_at = memparse(p+1, &p);
900 		e820_add_region(start_at, mem_size, E820_RESERVED);
901 	} else
902 		e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
903 
904 	return *p == '\0' ? 0 : -EINVAL;
905 }
906 early_param("memmap", parse_memmap_opt);
907 
908 void __init finish_e820_parsing(void)
909 {
910 	if (userdef) {
911 		u32 nr = e820.nr_map;
912 
913 		if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
914 			early_panic("Invalid user supplied memory map");
915 		e820.nr_map = nr;
916 
917 		printk(KERN_INFO "user-defined physical RAM map:\n");
918 		e820_print_map("user");
919 	}
920 }
921 
922 static inline const char *e820_type_to_string(int e820_type)
923 {
924 	switch (e820_type) {
925 	case E820_RESERVED_KERN:
926 	case E820_RAM:	return "System RAM";
927 	case E820_ACPI:	return "ACPI Tables";
928 	case E820_NVS:	return "ACPI Non-volatile Storage";
929 	case E820_UNUSABLE:	return "Unusable memory";
930 	default:	return "reserved";
931 	}
932 }
933 
934 /*
935  * Mark e820 reserved areas as busy for the resource manager.
936  */
937 static struct resource __initdata *e820_res;
938 void __init e820_reserve_resources(void)
939 {
940 	int i;
941 	struct resource *res;
942 	u64 end;
943 
944 	res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
945 	e820_res = res;
946 	for (i = 0; i < e820.nr_map; i++) {
947 		end = e820.map[i].addr + e820.map[i].size - 1;
948 		if (end != (resource_size_t)end) {
949 			res++;
950 			continue;
951 		}
952 		res->name = e820_type_to_string(e820.map[i].type);
953 		res->start = e820.map[i].addr;
954 		res->end = end;
955 
956 		res->flags = IORESOURCE_MEM;
957 
958 		/*
959 		 * don't register the region that could be conflicted with
960 		 * pci device BAR resource and insert them later in
961 		 * pcibios_resource_survey()
962 		 */
963 		if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20)) {
964 			res->flags |= IORESOURCE_BUSY;
965 			insert_resource(&iomem_resource, res);
966 		}
967 		res++;
968 	}
969 
970 	for (i = 0; i < e820_saved.nr_map; i++) {
971 		struct e820entry *entry = &e820_saved.map[i];
972 		firmware_map_add_early(entry->addr,
973 			entry->addr + entry->size - 1,
974 			e820_type_to_string(entry->type));
975 	}
976 }
977 
978 /* How much should we pad RAM ending depending on where it is? */
979 static unsigned long ram_alignment(resource_size_t pos)
980 {
981 	unsigned long mb = pos >> 20;
982 
983 	/* To 64kB in the first megabyte */
984 	if (!mb)
985 		return 64*1024;
986 
987 	/* To 1MB in the first 16MB */
988 	if (mb < 16)
989 		return 1024*1024;
990 
991 	/* To 64MB for anything above that */
992 	return 64*1024*1024;
993 }
994 
995 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
996 
997 void __init e820_reserve_resources_late(void)
998 {
999 	int i;
1000 	struct resource *res;
1001 
1002 	res = e820_res;
1003 	for (i = 0; i < e820.nr_map; i++) {
1004 		if (!res->parent && res->end)
1005 			insert_resource_expand_to_fit(&iomem_resource, res);
1006 		res++;
1007 	}
1008 
1009 	/*
1010 	 * Try to bump up RAM regions to reasonable boundaries to
1011 	 * avoid stolen RAM:
1012 	 */
1013 	for (i = 0; i < e820.nr_map; i++) {
1014 		struct e820entry *entry = &e820.map[i];
1015 		u64 start, end;
1016 
1017 		if (entry->type != E820_RAM)
1018 			continue;
1019 		start = entry->addr + entry->size;
1020 		end = round_up(start, ram_alignment(start)) - 1;
1021 		if (end > MAX_RESOURCE_SIZE)
1022 			end = MAX_RESOURCE_SIZE;
1023 		if (start >= end)
1024 			continue;
1025 		printk(KERN_DEBUG "reserve RAM buffer: %016llx - %016llx ",
1026 			       start, end);
1027 		reserve_region_with_split(&iomem_resource, start, end,
1028 					  "RAM buffer");
1029 	}
1030 }
1031 
1032 char *__init default_machine_specific_memory_setup(void)
1033 {
1034 	char *who = "BIOS-e820";
1035 	u32 new_nr;
1036 	/*
1037 	 * Try to copy the BIOS-supplied E820-map.
1038 	 *
1039 	 * Otherwise fake a memory map; one section from 0k->640k,
1040 	 * the next section from 1mb->appropriate_mem_k
1041 	 */
1042 	new_nr = boot_params.e820_entries;
1043 	sanitize_e820_map(boot_params.e820_map,
1044 			ARRAY_SIZE(boot_params.e820_map),
1045 			&new_nr);
1046 	boot_params.e820_entries = new_nr;
1047 	if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1048 	  < 0) {
1049 		u64 mem_size;
1050 
1051 		/* compare results from other methods and take the greater */
1052 		if (boot_params.alt_mem_k
1053 		    < boot_params.screen_info.ext_mem_k) {
1054 			mem_size = boot_params.screen_info.ext_mem_k;
1055 			who = "BIOS-88";
1056 		} else {
1057 			mem_size = boot_params.alt_mem_k;
1058 			who = "BIOS-e801";
1059 		}
1060 
1061 		e820.nr_map = 0;
1062 		e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1063 		e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1064 	}
1065 
1066 	/* In case someone cares... */
1067 	return who;
1068 }
1069 
1070 void __init setup_memory_map(void)
1071 {
1072 	char *who;
1073 
1074 	who = x86_init.resources.memory_setup();
1075 	memcpy(&e820_saved, &e820, sizeof(struct e820map));
1076 	printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1077 	e820_print_map(who);
1078 }
1079 
1080 void __init memblock_x86_fill(void)
1081 {
1082 	int i;
1083 	u64 end;
1084 
1085 	/*
1086 	 * EFI may have more than 128 entries
1087 	 * We are safe to enable resizing, beause memblock_x86_fill()
1088 	 * is rather later for x86
1089 	 */
1090 	memblock_can_resize = 1;
1091 
1092 	for (i = 0; i < e820.nr_map; i++) {
1093 		struct e820entry *ei = &e820.map[i];
1094 
1095 		end = ei->addr + ei->size;
1096 		if (end != (resource_size_t)end)
1097 			continue;
1098 
1099 		if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1100 			continue;
1101 
1102 		memblock_add(ei->addr, ei->size);
1103 	}
1104 
1105 	memblock_analyze();
1106 	memblock_dump_all();
1107 }
1108 
1109 void __init memblock_find_dma_reserve(void)
1110 {
1111 #ifdef CONFIG_X86_64
1112 	u64 free_size_pfn;
1113 	u64 mem_size_pfn;
1114 	/*
1115 	 * need to find out used area below MAX_DMA_PFN
1116 	 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1117 	 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1118 	 */
1119 	mem_size_pfn = memblock_x86_memory_in_range(0, MAX_DMA_PFN << PAGE_SHIFT) >> PAGE_SHIFT;
1120 	free_size_pfn = memblock_x86_free_memory_in_range(0, MAX_DMA_PFN << PAGE_SHIFT) >> PAGE_SHIFT;
1121 	set_dma_reserve(mem_size_pfn - free_size_pfn);
1122 #endif
1123 }
1124