xref: /linux/arch/x86/kernel/e820.c (revision a1ff5a7d78a036d6c2178ee5acd6ba4946243800)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Low level x86 E820 memory map handling functions.
4  *
5  * The firmware and bootloader passes us the "E820 table", which is the primary
6  * physical memory layout description available about x86 systems.
7  *
8  * The kernel takes the E820 memory layout and optionally modifies it with
9  * quirks and other tweaks, and feeds that into the generic Linux memory
10  * allocation code routines via a platform independent interface (memblock, etc.).
11  */
12 #include <linux/crash_dump.h>
13 #include <linux/memblock.h>
14 #include <linux/suspend.h>
15 #include <linux/acpi.h>
16 #include <linux/firmware-map.h>
17 #include <linux/sort.h>
18 #include <linux/memory_hotplug.h>
19 
20 #include <asm/e820/api.h>
21 #include <asm/setup.h>
22 
23 /*
24  * We organize the E820 table into three main data structures:
25  *
26  * - 'e820_table_firmware': the original firmware version passed to us by the
27  *   bootloader - not modified by the kernel. It is composed of two parts:
28  *   the first 128 E820 memory entries in boot_params.e820_table and the remaining
29  *   (if any) entries of the SETUP_E820_EXT nodes. We use this to:
30  *
31  *       - inform the user about the firmware's notion of memory layout
32  *         via /sys/firmware/memmap
33  *
34  *       - the hibernation code uses it to generate a kernel-independent CRC32
35  *         checksum of the physical memory layout of a system.
36  *
37  * - 'e820_table_kexec': a slightly modified (by the kernel) firmware version
38  *   passed to us by the bootloader - the major difference between
39  *   e820_table_firmware[] and this one is that, the latter marks the setup_data
40  *   list created by the EFI boot stub as reserved, so that kexec can reuse the
41  *   setup_data information in the second kernel. Besides, e820_table_kexec[]
42  *   might also be modified by the kexec itself to fake a mptable.
43  *   We use this to:
44  *
45  *       - kexec, which is a bootloader in disguise, uses the original E820
46  *         layout to pass to the kexec-ed kernel. This way the original kernel
47  *         can have a restricted E820 map while the kexec()-ed kexec-kernel
48  *         can have access to full memory - etc.
49  *
50  * - 'e820_table': this is the main E820 table that is massaged by the
51  *   low level x86 platform code, or modified by boot parameters, before
52  *   passed on to higher level MM layers.
53  *
54  * Once the E820 map has been converted to the standard Linux memory layout
55  * information its role stops - modifying it has no effect and does not get
56  * re-propagated. So its main role is a temporary bootstrap storage of firmware
57  * specific memory layout data during early bootup.
58  */
59 static struct e820_table e820_table_init		__initdata;
60 static struct e820_table e820_table_kexec_init		__initdata;
61 static struct e820_table e820_table_firmware_init	__initdata;
62 
63 struct e820_table *e820_table __refdata			= &e820_table_init;
64 struct e820_table *e820_table_kexec __refdata		= &e820_table_kexec_init;
65 struct e820_table *e820_table_firmware __refdata	= &e820_table_firmware_init;
66 
67 /* For PCI or other memory-mapped resources */
68 unsigned long pci_mem_start = 0xaeedbabe;
69 #ifdef CONFIG_PCI
70 EXPORT_SYMBOL(pci_mem_start);
71 #endif
72 
73 /*
74  * This function checks if any part of the range <start,end> is mapped
75  * with type.
76  */
_e820__mapped_any(struct e820_table * table,u64 start,u64 end,enum e820_type type)77 static bool _e820__mapped_any(struct e820_table *table,
78 			      u64 start, u64 end, enum e820_type type)
79 {
80 	int i;
81 
82 	for (i = 0; i < table->nr_entries; i++) {
83 		struct e820_entry *entry = &table->entries[i];
84 
85 		if (type && entry->type != type)
86 			continue;
87 		if (entry->addr >= end || entry->addr + entry->size <= start)
88 			continue;
89 		return true;
90 	}
91 	return false;
92 }
93 
e820__mapped_raw_any(u64 start,u64 end,enum e820_type type)94 bool e820__mapped_raw_any(u64 start, u64 end, enum e820_type type)
95 {
96 	return _e820__mapped_any(e820_table_firmware, start, end, type);
97 }
98 EXPORT_SYMBOL_GPL(e820__mapped_raw_any);
99 
e820__mapped_any(u64 start,u64 end,enum e820_type type)100 bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
101 {
102 	return _e820__mapped_any(e820_table, start, end, type);
103 }
104 EXPORT_SYMBOL_GPL(e820__mapped_any);
105 
106 /*
107  * This function checks if the entire <start,end> range is mapped with 'type'.
108  *
109  * Note: this function only works correctly once the E820 table is sorted and
110  * not-overlapping (at least for the range specified), which is the case normally.
111  */
__e820__mapped_all(u64 start,u64 end,enum e820_type type)112 static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
113 					     enum e820_type type)
114 {
115 	int i;
116 
117 	for (i = 0; i < e820_table->nr_entries; i++) {
118 		struct e820_entry *entry = &e820_table->entries[i];
119 
120 		if (type && entry->type != type)
121 			continue;
122 
123 		/* Is the region (part) in overlap with the current region? */
124 		if (entry->addr >= end || entry->addr + entry->size <= start)
125 			continue;
126 
127 		/*
128 		 * If the region is at the beginning of <start,end> we move
129 		 * 'start' to the end of the region since it's ok until there
130 		 */
131 		if (entry->addr <= start)
132 			start = entry->addr + entry->size;
133 
134 		/*
135 		 * If 'start' is now at or beyond 'end', we're done, full
136 		 * coverage of the desired range exists:
137 		 */
138 		if (start >= end)
139 			return entry;
140 	}
141 
142 	return NULL;
143 }
144 
145 /*
146  * This function checks if the entire range <start,end> is mapped with type.
147  */
e820__mapped_all(u64 start,u64 end,enum e820_type type)148 bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
149 {
150 	return __e820__mapped_all(start, end, type);
151 }
152 
153 /*
154  * This function returns the type associated with the range <start,end>.
155  */
e820__get_entry_type(u64 start,u64 end)156 int e820__get_entry_type(u64 start, u64 end)
157 {
158 	struct e820_entry *entry = __e820__mapped_all(start, end, 0);
159 
160 	return entry ? entry->type : -EINVAL;
161 }
162 
163 /*
164  * Add a memory region to the kernel E820 map.
165  */
__e820__range_add(struct e820_table * table,u64 start,u64 size,enum e820_type type)166 static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
167 {
168 	int x = table->nr_entries;
169 
170 	if (x >= ARRAY_SIZE(table->entries)) {
171 		pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
172 		       start, start + size - 1);
173 		return;
174 	}
175 
176 	table->entries[x].addr = start;
177 	table->entries[x].size = size;
178 	table->entries[x].type = type;
179 	table->nr_entries++;
180 }
181 
e820__range_add(u64 start,u64 size,enum e820_type type)182 void __init e820__range_add(u64 start, u64 size, enum e820_type type)
183 {
184 	__e820__range_add(e820_table, start, size, type);
185 }
186 
e820_print_type(enum e820_type type)187 static void __init e820_print_type(enum e820_type type)
188 {
189 	switch (type) {
190 	case E820_TYPE_RAM:		/* Fall through: */
191 	case E820_TYPE_RESERVED_KERN:	pr_cont("usable");			break;
192 	case E820_TYPE_RESERVED:	pr_cont("reserved");			break;
193 	case E820_TYPE_SOFT_RESERVED:	pr_cont("soft reserved");		break;
194 	case E820_TYPE_ACPI:		pr_cont("ACPI data");			break;
195 	case E820_TYPE_NVS:		pr_cont("ACPI NVS");			break;
196 	case E820_TYPE_UNUSABLE:	pr_cont("unusable");			break;
197 	case E820_TYPE_PMEM:		/* Fall through: */
198 	case E820_TYPE_PRAM:		pr_cont("persistent (type %u)", type);	break;
199 	default:			pr_cont("type %u", type);		break;
200 	}
201 }
202 
e820__print_table(char * who)203 void __init e820__print_table(char *who)
204 {
205 	int i;
206 
207 	for (i = 0; i < e820_table->nr_entries; i++) {
208 		pr_info("%s: [mem %#018Lx-%#018Lx] ",
209 			who,
210 			e820_table->entries[i].addr,
211 			e820_table->entries[i].addr + e820_table->entries[i].size - 1);
212 
213 		e820_print_type(e820_table->entries[i].type);
214 		pr_cont("\n");
215 	}
216 }
217 
218 /*
219  * Sanitize an E820 map.
220  *
221  * Some E820 layouts include overlapping entries. The following
222  * replaces the original E820 map with a new one, removing overlaps,
223  * and resolving conflicting memory types in favor of highest
224  * numbered type.
225  *
226  * The input parameter 'entries' points to an array of 'struct
227  * e820_entry' which on entry has elements in the range [0, *nr_entries)
228  * valid, and which has space for up to max_nr_entries entries.
229  * On return, the resulting sanitized E820 map entries will be in
230  * overwritten in the same location, starting at 'entries'.
231  *
232  * The integer pointed to by nr_entries must be valid on entry (the
233  * current number of valid entries located at 'entries'). If the
234  * sanitizing succeeds the *nr_entries will be updated with the new
235  * number of valid entries (something no more than max_nr_entries).
236  *
237  * The return value from e820__update_table() is zero if it
238  * successfully 'sanitized' the map entries passed in, and is -1
239  * if it did nothing, which can happen if either of (1) it was
240  * only passed one map entry, or (2) any of the input map entries
241  * were invalid (start + size < start, meaning that the size was
242  * so big the described memory range wrapped around through zero.)
243  *
244  *	Visually we're performing the following
245  *	(1,2,3,4 = memory types)...
246  *
247  *	Sample memory map (w/overlaps):
248  *	   ____22__________________
249  *	   ______________________4_
250  *	   ____1111________________
251  *	   _44_____________________
252  *	   11111111________________
253  *	   ____________________33__
254  *	   ___________44___________
255  *	   __________33333_________
256  *	   ______________22________
257  *	   ___________________2222_
258  *	   _________111111111______
259  *	   _____________________11_
260  *	   _________________4______
261  *
262  *	Sanitized equivalent (no overlap):
263  *	   1_______________________
264  *	   _44_____________________
265  *	   ___1____________________
266  *	   ____22__________________
267  *	   ______11________________
268  *	   _________1______________
269  *	   __________3_____________
270  *	   ___________44___________
271  *	   _____________33_________
272  *	   _______________2________
273  *	   ________________1_______
274  *	   _________________4______
275  *	   ___________________2____
276  *	   ____________________33__
277  *	   ______________________4_
278  */
279 struct change_member {
280 	/* Pointer to the original entry: */
281 	struct e820_entry	*entry;
282 	/* Address for this change point: */
283 	unsigned long long	addr;
284 };
285 
286 static struct change_member	change_point_list[2*E820_MAX_ENTRIES]	__initdata;
287 static struct change_member	*change_point[2*E820_MAX_ENTRIES]	__initdata;
288 static struct e820_entry	*overlap_list[E820_MAX_ENTRIES]		__initdata;
289 static struct e820_entry	new_entries[E820_MAX_ENTRIES]		__initdata;
290 
cpcompare(const void * a,const void * b)291 static int __init cpcompare(const void *a, const void *b)
292 {
293 	struct change_member * const *app = a, * const *bpp = b;
294 	const struct change_member *ap = *app, *bp = *bpp;
295 
296 	/*
297 	 * Inputs are pointers to two elements of change_point[].  If their
298 	 * addresses are not equal, their difference dominates.  If the addresses
299 	 * are equal, then consider one that represents the end of its region
300 	 * to be greater than one that does not.
301 	 */
302 	if (ap->addr != bp->addr)
303 		return ap->addr > bp->addr ? 1 : -1;
304 
305 	return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
306 }
307 
e820_nomerge(enum e820_type type)308 static bool e820_nomerge(enum e820_type type)
309 {
310 	/*
311 	 * These types may indicate distinct platform ranges aligned to
312 	 * numa node, protection domain, performance domain, or other
313 	 * boundaries. Do not merge them.
314 	 */
315 	if (type == E820_TYPE_PRAM)
316 		return true;
317 	if (type == E820_TYPE_SOFT_RESERVED)
318 		return true;
319 	return false;
320 }
321 
e820__update_table(struct e820_table * table)322 int __init e820__update_table(struct e820_table *table)
323 {
324 	struct e820_entry *entries = table->entries;
325 	u32 max_nr_entries = ARRAY_SIZE(table->entries);
326 	enum e820_type current_type, last_type;
327 	unsigned long long last_addr;
328 	u32 new_nr_entries, overlap_entries;
329 	u32 i, chg_idx, chg_nr;
330 
331 	/* If there's only one memory region, don't bother: */
332 	if (table->nr_entries < 2)
333 		return -1;
334 
335 	BUG_ON(table->nr_entries > max_nr_entries);
336 
337 	/* Bail out if we find any unreasonable addresses in the map: */
338 	for (i = 0; i < table->nr_entries; i++) {
339 		if (entries[i].addr + entries[i].size < entries[i].addr)
340 			return -1;
341 	}
342 
343 	/* Create pointers for initial change-point information (for sorting): */
344 	for (i = 0; i < 2 * table->nr_entries; i++)
345 		change_point[i] = &change_point_list[i];
346 
347 	/*
348 	 * Record all known change-points (starting and ending addresses),
349 	 * omitting empty memory regions:
350 	 */
351 	chg_idx = 0;
352 	for (i = 0; i < table->nr_entries; i++)	{
353 		if (entries[i].size != 0) {
354 			change_point[chg_idx]->addr	= entries[i].addr;
355 			change_point[chg_idx++]->entry	= &entries[i];
356 			change_point[chg_idx]->addr	= entries[i].addr + entries[i].size;
357 			change_point[chg_idx++]->entry	= &entries[i];
358 		}
359 	}
360 	chg_nr = chg_idx;
361 
362 	/* Sort change-point list by memory addresses (low -> high): */
363 	sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
364 
365 	/* Create a new memory map, removing overlaps: */
366 	overlap_entries = 0;	 /* Number of entries in the overlap table */
367 	new_nr_entries = 0;	 /* Index for creating new map entries */
368 	last_type = 0;		 /* Start with undefined memory type */
369 	last_addr = 0;		 /* Start with 0 as last starting address */
370 
371 	/* Loop through change-points, determining effect on the new map: */
372 	for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
373 		/* Keep track of all overlapping entries */
374 		if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
375 			/* Add map entry to overlap list (> 1 entry implies an overlap) */
376 			overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
377 		} else {
378 			/* Remove entry from list (order independent, so swap with last): */
379 			for (i = 0; i < overlap_entries; i++) {
380 				if (overlap_list[i] == change_point[chg_idx]->entry)
381 					overlap_list[i] = overlap_list[overlap_entries-1];
382 			}
383 			overlap_entries--;
384 		}
385 		/*
386 		 * If there are overlapping entries, decide which
387 		 * "type" to use (larger value takes precedence --
388 		 * 1=usable, 2,3,4,4+=unusable)
389 		 */
390 		current_type = 0;
391 		for (i = 0; i < overlap_entries; i++) {
392 			if (overlap_list[i]->type > current_type)
393 				current_type = overlap_list[i]->type;
394 		}
395 
396 		/* Continue building up new map based on this information: */
397 		if (current_type != last_type || e820_nomerge(current_type)) {
398 			if (last_type) {
399 				new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
400 				/* Move forward only if the new size was non-zero: */
401 				if (new_entries[new_nr_entries].size != 0)
402 					/* No more space left for new entries? */
403 					if (++new_nr_entries >= max_nr_entries)
404 						break;
405 			}
406 			if (current_type) {
407 				new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
408 				new_entries[new_nr_entries].type = current_type;
409 				last_addr = change_point[chg_idx]->addr;
410 			}
411 			last_type = current_type;
412 		}
413 	}
414 
415 	/* Copy the new entries into the original location: */
416 	memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
417 	table->nr_entries = new_nr_entries;
418 
419 	return 0;
420 }
421 
__append_e820_table(struct boot_e820_entry * entries,u32 nr_entries)422 static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
423 {
424 	struct boot_e820_entry *entry = entries;
425 
426 	while (nr_entries) {
427 		u64 start = entry->addr;
428 		u64 size = entry->size;
429 		u64 end = start + size - 1;
430 		u32 type = entry->type;
431 
432 		/* Ignore the entry on 64-bit overflow: */
433 		if (start > end && likely(size))
434 			return -1;
435 
436 		e820__range_add(start, size, type);
437 
438 		entry++;
439 		nr_entries--;
440 	}
441 	return 0;
442 }
443 
444 /*
445  * Copy the BIOS E820 map into a safe place.
446  *
447  * Sanity-check it while we're at it..
448  *
449  * If we're lucky and live on a modern system, the setup code
450  * will have given us a memory map that we can use to properly
451  * set up memory.  If we aren't, we'll fake a memory map.
452  */
append_e820_table(struct boot_e820_entry * entries,u32 nr_entries)453 static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
454 {
455 	/* Only one memory region (or negative)? Ignore it */
456 	if (nr_entries < 2)
457 		return -1;
458 
459 	return __append_e820_table(entries, nr_entries);
460 }
461 
462 static u64 __init
__e820__range_update(struct e820_table * table,u64 start,u64 size,enum e820_type old_type,enum e820_type new_type)463 __e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
464 {
465 	u64 end;
466 	unsigned int i;
467 	u64 real_updated_size = 0;
468 
469 	BUG_ON(old_type == new_type);
470 
471 	if (size > (ULLONG_MAX - start))
472 		size = ULLONG_MAX - start;
473 
474 	end = start + size;
475 	printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
476 	e820_print_type(old_type);
477 	pr_cont(" ==> ");
478 	e820_print_type(new_type);
479 	pr_cont("\n");
480 
481 	for (i = 0; i < table->nr_entries; i++) {
482 		struct e820_entry *entry = &table->entries[i];
483 		u64 final_start, final_end;
484 		u64 entry_end;
485 
486 		if (entry->type != old_type)
487 			continue;
488 
489 		entry_end = entry->addr + entry->size;
490 
491 		/* Completely covered by new range? */
492 		if (entry->addr >= start && entry_end <= end) {
493 			entry->type = new_type;
494 			real_updated_size += entry->size;
495 			continue;
496 		}
497 
498 		/* New range is completely covered? */
499 		if (entry->addr < start && entry_end > end) {
500 			__e820__range_add(table, start, size, new_type);
501 			__e820__range_add(table, end, entry_end - end, entry->type);
502 			entry->size = start - entry->addr;
503 			real_updated_size += size;
504 			continue;
505 		}
506 
507 		/* Partially covered: */
508 		final_start = max(start, entry->addr);
509 		final_end = min(end, entry_end);
510 		if (final_start >= final_end)
511 			continue;
512 
513 		__e820__range_add(table, final_start, final_end - final_start, new_type);
514 
515 		real_updated_size += final_end - final_start;
516 
517 		/*
518 		 * Left range could be head or tail, so need to update
519 		 * its size first:
520 		 */
521 		entry->size -= final_end - final_start;
522 		if (entry->addr < final_start)
523 			continue;
524 
525 		entry->addr = final_end;
526 	}
527 	return real_updated_size;
528 }
529 
e820__range_update(u64 start,u64 size,enum e820_type old_type,enum e820_type new_type)530 u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
531 {
532 	return __e820__range_update(e820_table, start, size, old_type, new_type);
533 }
534 
e820__range_update_table(struct e820_table * t,u64 start,u64 size,enum e820_type old_type,enum e820_type new_type)535 u64 __init e820__range_update_table(struct e820_table *t, u64 start, u64 size,
536 				    enum e820_type old_type, enum e820_type new_type)
537 {
538 	return __e820__range_update(t, start, size, old_type, new_type);
539 }
540 
541 /* Remove a range of memory from the E820 table: */
e820__range_remove(u64 start,u64 size,enum e820_type old_type,bool check_type)542 u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
543 {
544 	int i;
545 	u64 end;
546 	u64 real_removed_size = 0;
547 
548 	if (size > (ULLONG_MAX - start))
549 		size = ULLONG_MAX - start;
550 
551 	end = start + size;
552 	printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
553 	if (check_type)
554 		e820_print_type(old_type);
555 	pr_cont("\n");
556 
557 	for (i = 0; i < e820_table->nr_entries; i++) {
558 		struct e820_entry *entry = &e820_table->entries[i];
559 		u64 final_start, final_end;
560 		u64 entry_end;
561 
562 		if (check_type && entry->type != old_type)
563 			continue;
564 
565 		entry_end = entry->addr + entry->size;
566 
567 		/* Completely covered? */
568 		if (entry->addr >= start && entry_end <= end) {
569 			real_removed_size += entry->size;
570 			memset(entry, 0, sizeof(*entry));
571 			continue;
572 		}
573 
574 		/* Is the new range completely covered? */
575 		if (entry->addr < start && entry_end > end) {
576 			e820__range_add(end, entry_end - end, entry->type);
577 			entry->size = start - entry->addr;
578 			real_removed_size += size;
579 			continue;
580 		}
581 
582 		/* Partially covered: */
583 		final_start = max(start, entry->addr);
584 		final_end = min(end, entry_end);
585 		if (final_start >= final_end)
586 			continue;
587 
588 		real_removed_size += final_end - final_start;
589 
590 		/*
591 		 * Left range could be head or tail, so need to update
592 		 * the size first:
593 		 */
594 		entry->size -= final_end - final_start;
595 		if (entry->addr < final_start)
596 			continue;
597 
598 		entry->addr = final_end;
599 	}
600 	return real_removed_size;
601 }
602 
e820__update_table_print(void)603 void __init e820__update_table_print(void)
604 {
605 	if (e820__update_table(e820_table))
606 		return;
607 
608 	pr_info("modified physical RAM map:\n");
609 	e820__print_table("modified");
610 }
611 
e820__update_table_kexec(void)612 static void __init e820__update_table_kexec(void)
613 {
614 	e820__update_table(e820_table_kexec);
615 }
616 
617 #define MAX_GAP_END 0x100000000ull
618 
619 /*
620  * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
621  */
e820_search_gap(unsigned long * gapstart,unsigned long * gapsize)622 static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
623 {
624 	unsigned long long last = MAX_GAP_END;
625 	int i = e820_table->nr_entries;
626 	int found = 0;
627 
628 	while (--i >= 0) {
629 		unsigned long long start = e820_table->entries[i].addr;
630 		unsigned long long end = start + e820_table->entries[i].size;
631 
632 		/*
633 		 * Since "last" is at most 4GB, we know we'll
634 		 * fit in 32 bits if this condition is true:
635 		 */
636 		if (last > end) {
637 			unsigned long gap = last - end;
638 
639 			if (gap >= *gapsize) {
640 				*gapsize = gap;
641 				*gapstart = end;
642 				found = 1;
643 			}
644 		}
645 		if (start < last)
646 			last = start;
647 	}
648 	return found;
649 }
650 
651 /*
652  * Search for the biggest gap in the low 32 bits of the E820
653  * memory space. We pass this space to the PCI subsystem, so
654  * that it can assign MMIO resources for hotplug or
655  * unconfigured devices in.
656  *
657  * Hopefully the BIOS let enough space left.
658  */
e820__setup_pci_gap(void)659 __init void e820__setup_pci_gap(void)
660 {
661 	unsigned long gapstart, gapsize;
662 	int found;
663 
664 	gapsize = 0x400000;
665 	found  = e820_search_gap(&gapstart, &gapsize);
666 
667 	if (!found) {
668 #ifdef CONFIG_X86_64
669 		gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
670 		pr_err("Cannot find an available gap in the 32-bit address range\n");
671 		pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
672 #else
673 		gapstart = 0x10000000;
674 #endif
675 	}
676 
677 	/*
678 	 * e820__reserve_resources_late() protects stolen RAM already:
679 	 */
680 	pci_mem_start = gapstart;
681 
682 	pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
683 		gapstart, gapstart + gapsize - 1);
684 }
685 
686 /*
687  * Called late during init, in free_initmem().
688  *
689  * Initial e820_table and e820_table_kexec are largish __initdata arrays.
690  *
691  * Copy them to a (usually much smaller) dynamically allocated area that is
692  * sized precisely after the number of e820 entries.
693  *
694  * This is done after we've performed all the fixes and tweaks to the tables.
695  * All functions which modify them are __init functions, which won't exist
696  * after free_initmem().
697  */
e820__reallocate_tables(void)698 __init void e820__reallocate_tables(void)
699 {
700 	struct e820_table *n;
701 	int size;
702 
703 	size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
704 	n = kmemdup(e820_table, size, GFP_KERNEL);
705 	BUG_ON(!n);
706 	e820_table = n;
707 
708 	size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
709 	n = kmemdup(e820_table_kexec, size, GFP_KERNEL);
710 	BUG_ON(!n);
711 	e820_table_kexec = n;
712 
713 	size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
714 	n = kmemdup(e820_table_firmware, size, GFP_KERNEL);
715 	BUG_ON(!n);
716 	e820_table_firmware = n;
717 }
718 
719 /*
720  * Because of the small fixed size of struct boot_params, only the first
721  * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
722  * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
723  * struct setup_data, which is parsed here.
724  */
e820__memory_setup_extended(u64 phys_addr,u32 data_len)725 void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
726 {
727 	int entries;
728 	struct boot_e820_entry *extmap;
729 	struct setup_data *sdata;
730 
731 	sdata = early_memremap(phys_addr, data_len);
732 	entries = sdata->len / sizeof(*extmap);
733 	extmap = (struct boot_e820_entry *)(sdata->data);
734 
735 	__append_e820_table(extmap, entries);
736 	e820__update_table(e820_table);
737 
738 	memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
739 	memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
740 
741 	early_memunmap(sdata, data_len);
742 	pr_info("extended physical RAM map:\n");
743 	e820__print_table("extended");
744 }
745 
746 /*
747  * Find the ranges of physical addresses that do not correspond to
748  * E820 RAM areas and register the corresponding pages as 'nosave' for
749  * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
750  *
751  * This function requires the E820 map to be sorted and without any
752  * overlapping entries.
753  */
e820__register_nosave_regions(unsigned long limit_pfn)754 void __init e820__register_nosave_regions(unsigned long limit_pfn)
755 {
756 	int i;
757 	unsigned long pfn = 0;
758 
759 	for (i = 0; i < e820_table->nr_entries; i++) {
760 		struct e820_entry *entry = &e820_table->entries[i];
761 
762 		if (pfn < PFN_UP(entry->addr))
763 			register_nosave_region(pfn, PFN_UP(entry->addr));
764 
765 		pfn = PFN_DOWN(entry->addr + entry->size);
766 
767 		if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
768 			register_nosave_region(PFN_UP(entry->addr), pfn);
769 
770 		if (pfn >= limit_pfn)
771 			break;
772 	}
773 }
774 
775 #ifdef CONFIG_ACPI
776 /*
777  * Register ACPI NVS memory regions, so that we can save/restore them during
778  * hibernation and the subsequent resume:
779  */
e820__register_nvs_regions(void)780 static int __init e820__register_nvs_regions(void)
781 {
782 	int i;
783 
784 	for (i = 0; i < e820_table->nr_entries; i++) {
785 		struct e820_entry *entry = &e820_table->entries[i];
786 
787 		if (entry->type == E820_TYPE_NVS)
788 			acpi_nvs_register(entry->addr, entry->size);
789 	}
790 
791 	return 0;
792 }
793 core_initcall(e820__register_nvs_regions);
794 #endif
795 
796 /*
797  * Allocate the requested number of bytes with the requested alignment
798  * and return (the physical address) to the caller. Also register this
799  * range in the 'kexec' E820 table as a reserved range.
800  *
801  * This allows kexec to fake a new mptable, as if it came from the real
802  * system.
803  */
e820__memblock_alloc_reserved(u64 size,u64 align)804 u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
805 {
806 	u64 addr;
807 
808 	addr = memblock_phys_alloc(size, align);
809 	if (addr) {
810 		e820__range_update_table(e820_table_kexec, addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
811 		pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
812 		e820__update_table_kexec();
813 	}
814 
815 	return addr;
816 }
817 
818 #ifdef CONFIG_X86_32
819 # ifdef CONFIG_X86_PAE
820 #  define MAX_ARCH_PFN		(1ULL<<(36-PAGE_SHIFT))
821 # else
822 #  define MAX_ARCH_PFN		(1ULL<<(32-PAGE_SHIFT))
823 # endif
824 #else /* CONFIG_X86_32 */
825 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
826 #endif
827 
828 /*
829  * Find the highest page frame number we have available
830  */
e820__end_ram_pfn(unsigned long limit_pfn)831 static unsigned long __init e820__end_ram_pfn(unsigned long limit_pfn)
832 {
833 	int i;
834 	unsigned long last_pfn = 0;
835 	unsigned long max_arch_pfn = MAX_ARCH_PFN;
836 
837 	for (i = 0; i < e820_table->nr_entries; i++) {
838 		struct e820_entry *entry = &e820_table->entries[i];
839 		unsigned long start_pfn;
840 		unsigned long end_pfn;
841 
842 		if (entry->type != E820_TYPE_RAM &&
843 		    entry->type != E820_TYPE_ACPI)
844 			continue;
845 
846 		start_pfn = entry->addr >> PAGE_SHIFT;
847 		end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
848 
849 		if (start_pfn >= limit_pfn)
850 			continue;
851 		if (end_pfn > limit_pfn) {
852 			last_pfn = limit_pfn;
853 			break;
854 		}
855 		if (end_pfn > last_pfn)
856 			last_pfn = end_pfn;
857 	}
858 
859 	if (last_pfn > max_arch_pfn)
860 		last_pfn = max_arch_pfn;
861 
862 	pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
863 		last_pfn, max_arch_pfn);
864 	return last_pfn;
865 }
866 
e820__end_of_ram_pfn(void)867 unsigned long __init e820__end_of_ram_pfn(void)
868 {
869 	return e820__end_ram_pfn(MAX_ARCH_PFN);
870 }
871 
e820__end_of_low_ram_pfn(void)872 unsigned long __init e820__end_of_low_ram_pfn(void)
873 {
874 	return e820__end_ram_pfn(1UL << (32 - PAGE_SHIFT));
875 }
876 
early_panic(char * msg)877 static void __init early_panic(char *msg)
878 {
879 	early_printk(msg);
880 	panic(msg);
881 }
882 
883 static int userdef __initdata;
884 
885 /* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
parse_memopt(char * p)886 static int __init parse_memopt(char *p)
887 {
888 	u64 mem_size;
889 
890 	if (!p)
891 		return -EINVAL;
892 
893 	if (!strcmp(p, "nopentium")) {
894 #ifdef CONFIG_X86_32
895 		setup_clear_cpu_cap(X86_FEATURE_PSE);
896 		return 0;
897 #else
898 		pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
899 		return -EINVAL;
900 #endif
901 	}
902 
903 	userdef = 1;
904 	mem_size = memparse(p, &p);
905 
906 	/* Don't remove all memory when getting "mem={invalid}" parameter: */
907 	if (mem_size == 0)
908 		return -EINVAL;
909 
910 	e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
911 
912 #ifdef CONFIG_MEMORY_HOTPLUG
913 	max_mem_size = mem_size;
914 #endif
915 
916 	return 0;
917 }
918 early_param("mem", parse_memopt);
919 
parse_memmap_one(char * p)920 static int __init parse_memmap_one(char *p)
921 {
922 	char *oldp;
923 	u64 start_at, mem_size;
924 
925 	if (!p)
926 		return -EINVAL;
927 
928 	if (!strncmp(p, "exactmap", 8)) {
929 		e820_table->nr_entries = 0;
930 		userdef = 1;
931 		return 0;
932 	}
933 
934 	oldp = p;
935 	mem_size = memparse(p, &p);
936 	if (p == oldp)
937 		return -EINVAL;
938 
939 	userdef = 1;
940 	if (*p == '@') {
941 		start_at = memparse(p+1, &p);
942 		e820__range_add(start_at, mem_size, E820_TYPE_RAM);
943 	} else if (*p == '#') {
944 		start_at = memparse(p+1, &p);
945 		e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
946 	} else if (*p == '$') {
947 		start_at = memparse(p+1, &p);
948 		e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
949 	} else if (*p == '!') {
950 		start_at = memparse(p+1, &p);
951 		e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
952 	} else if (*p == '%') {
953 		enum e820_type from = 0, to = 0;
954 
955 		start_at = memparse(p + 1, &p);
956 		if (*p == '-')
957 			from = simple_strtoull(p + 1, &p, 0);
958 		if (*p == '+')
959 			to = simple_strtoull(p + 1, &p, 0);
960 		if (*p != '\0')
961 			return -EINVAL;
962 		if (from && to)
963 			e820__range_update(start_at, mem_size, from, to);
964 		else if (to)
965 			e820__range_add(start_at, mem_size, to);
966 		else if (from)
967 			e820__range_remove(start_at, mem_size, from, 1);
968 		else
969 			e820__range_remove(start_at, mem_size, 0, 0);
970 	} else {
971 		e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
972 	}
973 
974 	return *p == '\0' ? 0 : -EINVAL;
975 }
976 
parse_memmap_opt(char * str)977 static int __init parse_memmap_opt(char *str)
978 {
979 	while (str) {
980 		char *k = strchr(str, ',');
981 
982 		if (k)
983 			*k++ = 0;
984 
985 		parse_memmap_one(str);
986 		str = k;
987 	}
988 
989 	return 0;
990 }
991 early_param("memmap", parse_memmap_opt);
992 
993 /*
994  * Reserve all entries from the bootloader's extensible data nodes list,
995  * because if present we are going to use it later on to fetch e820
996  * entries from it:
997  */
e820__reserve_setup_data(void)998 void __init e820__reserve_setup_data(void)
999 {
1000 	struct setup_indirect *indirect;
1001 	struct setup_data *data;
1002 	u64 pa_data, pa_next;
1003 	u32 len;
1004 
1005 	pa_data = boot_params.hdr.setup_data;
1006 	if (!pa_data)
1007 		return;
1008 
1009 	while (pa_data) {
1010 		data = early_memremap(pa_data, sizeof(*data));
1011 		if (!data) {
1012 			pr_warn("e820: failed to memremap setup_data entry\n");
1013 			return;
1014 		}
1015 
1016 		len = sizeof(*data);
1017 		pa_next = data->next;
1018 
1019 		e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1020 
1021 		if (data->type == SETUP_INDIRECT) {
1022 			len += data->len;
1023 			early_memunmap(data, sizeof(*data));
1024 			data = early_memremap(pa_data, len);
1025 			if (!data) {
1026 				pr_warn("e820: failed to memremap indirect setup_data\n");
1027 				return;
1028 			}
1029 
1030 			indirect = (struct setup_indirect *)data->data;
1031 
1032 			if (indirect->type != SETUP_INDIRECT)
1033 				e820__range_update(indirect->addr, indirect->len,
1034 						   E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
1035 		}
1036 
1037 		pa_data = pa_next;
1038 		early_memunmap(data, len);
1039 	}
1040 
1041 	e820__update_table(e820_table);
1042 
1043 	pr_info("extended physical RAM map:\n");
1044 	e820__print_table("reserve setup_data");
1045 }
1046 
1047 /*
1048  * Called after parse_early_param(), after early parameters (such as mem=)
1049  * have been processed, in which case we already have an E820 table filled in
1050  * via the parameter callback function(s), but it's not sorted and printed yet:
1051  */
e820__finish_early_params(void)1052 void __init e820__finish_early_params(void)
1053 {
1054 	if (userdef) {
1055 		if (e820__update_table(e820_table) < 0)
1056 			early_panic("Invalid user supplied memory map");
1057 
1058 		pr_info("user-defined physical RAM map:\n");
1059 		e820__print_table("user");
1060 	}
1061 }
1062 
e820_type_to_string(struct e820_entry * entry)1063 static const char *__init e820_type_to_string(struct e820_entry *entry)
1064 {
1065 	switch (entry->type) {
1066 	case E820_TYPE_RESERVED_KERN:	/* Fall-through: */
1067 	case E820_TYPE_RAM:		return "System RAM";
1068 	case E820_TYPE_ACPI:		return "ACPI Tables";
1069 	case E820_TYPE_NVS:		return "ACPI Non-volatile Storage";
1070 	case E820_TYPE_UNUSABLE:	return "Unusable memory";
1071 	case E820_TYPE_PRAM:		return "Persistent Memory (legacy)";
1072 	case E820_TYPE_PMEM:		return "Persistent Memory";
1073 	case E820_TYPE_RESERVED:	return "Reserved";
1074 	case E820_TYPE_SOFT_RESERVED:	return "Soft Reserved";
1075 	default:			return "Unknown E820 type";
1076 	}
1077 }
1078 
e820_type_to_iomem_type(struct e820_entry * entry)1079 static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1080 {
1081 	switch (entry->type) {
1082 	case E820_TYPE_RESERVED_KERN:	/* Fall-through: */
1083 	case E820_TYPE_RAM:		return IORESOURCE_SYSTEM_RAM;
1084 	case E820_TYPE_ACPI:		/* Fall-through: */
1085 	case E820_TYPE_NVS:		/* Fall-through: */
1086 	case E820_TYPE_UNUSABLE:	/* Fall-through: */
1087 	case E820_TYPE_PRAM:		/* Fall-through: */
1088 	case E820_TYPE_PMEM:		/* Fall-through: */
1089 	case E820_TYPE_RESERVED:	/* Fall-through: */
1090 	case E820_TYPE_SOFT_RESERVED:	/* Fall-through: */
1091 	default:			return IORESOURCE_MEM;
1092 	}
1093 }
1094 
e820_type_to_iores_desc(struct e820_entry * entry)1095 static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1096 {
1097 	switch (entry->type) {
1098 	case E820_TYPE_ACPI:		return IORES_DESC_ACPI_TABLES;
1099 	case E820_TYPE_NVS:		return IORES_DESC_ACPI_NV_STORAGE;
1100 	case E820_TYPE_PMEM:		return IORES_DESC_PERSISTENT_MEMORY;
1101 	case E820_TYPE_PRAM:		return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1102 	case E820_TYPE_RESERVED:	return IORES_DESC_RESERVED;
1103 	case E820_TYPE_SOFT_RESERVED:	return IORES_DESC_SOFT_RESERVED;
1104 	case E820_TYPE_RESERVED_KERN:	/* Fall-through: */
1105 	case E820_TYPE_RAM:		/* Fall-through: */
1106 	case E820_TYPE_UNUSABLE:	/* Fall-through: */
1107 	default:			return IORES_DESC_NONE;
1108 	}
1109 }
1110 
do_mark_busy(enum e820_type type,struct resource * res)1111 static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1112 {
1113 	/* this is the legacy bios/dos rom-shadow + mmio region */
1114 	if (res->start < (1ULL<<20))
1115 		return true;
1116 
1117 	/*
1118 	 * Treat persistent memory and other special memory ranges like
1119 	 * device memory, i.e. reserve it for exclusive use of a driver
1120 	 */
1121 	switch (type) {
1122 	case E820_TYPE_RESERVED:
1123 	case E820_TYPE_SOFT_RESERVED:
1124 	case E820_TYPE_PRAM:
1125 	case E820_TYPE_PMEM:
1126 		return false;
1127 	case E820_TYPE_RESERVED_KERN:
1128 	case E820_TYPE_RAM:
1129 	case E820_TYPE_ACPI:
1130 	case E820_TYPE_NVS:
1131 	case E820_TYPE_UNUSABLE:
1132 	default:
1133 		return true;
1134 	}
1135 }
1136 
1137 /*
1138  * Mark E820 reserved areas as busy for the resource manager:
1139  */
1140 
1141 static struct resource __initdata *e820_res;
1142 
e820__reserve_resources(void)1143 void __init e820__reserve_resources(void)
1144 {
1145 	int i;
1146 	struct resource *res;
1147 	u64 end;
1148 
1149 	res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1150 			     SMP_CACHE_BYTES);
1151 	if (!res)
1152 		panic("%s: Failed to allocate %zu bytes\n", __func__,
1153 		      sizeof(*res) * e820_table->nr_entries);
1154 	e820_res = res;
1155 
1156 	for (i = 0; i < e820_table->nr_entries; i++) {
1157 		struct e820_entry *entry = e820_table->entries + i;
1158 
1159 		end = entry->addr + entry->size - 1;
1160 		if (end != (resource_size_t)end) {
1161 			res++;
1162 			continue;
1163 		}
1164 		res->start = entry->addr;
1165 		res->end   = end;
1166 		res->name  = e820_type_to_string(entry);
1167 		res->flags = e820_type_to_iomem_type(entry);
1168 		res->desc  = e820_type_to_iores_desc(entry);
1169 
1170 		/*
1171 		 * Don't register the region that could be conflicted with
1172 		 * PCI device BAR resources and insert them later in
1173 		 * pcibios_resource_survey():
1174 		 */
1175 		if (do_mark_busy(entry->type, res)) {
1176 			res->flags |= IORESOURCE_BUSY;
1177 			insert_resource(&iomem_resource, res);
1178 		}
1179 		res++;
1180 	}
1181 
1182 	/* Expose the bootloader-provided memory layout to the sysfs. */
1183 	for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1184 		struct e820_entry *entry = e820_table_firmware->entries + i;
1185 
1186 		firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1187 	}
1188 }
1189 
1190 /*
1191  * How much should we pad the end of RAM, depending on where it is?
1192  */
ram_alignment(resource_size_t pos)1193 static unsigned long __init ram_alignment(resource_size_t pos)
1194 {
1195 	unsigned long mb = pos >> 20;
1196 
1197 	/* To 64kB in the first megabyte */
1198 	if (!mb)
1199 		return 64*1024;
1200 
1201 	/* To 1MB in the first 16MB */
1202 	if (mb < 16)
1203 		return 1024*1024;
1204 
1205 	/* To 64MB for anything above that */
1206 	return 64*1024*1024;
1207 }
1208 
1209 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1210 
e820__reserve_resources_late(void)1211 void __init e820__reserve_resources_late(void)
1212 {
1213 	int i;
1214 	struct resource *res;
1215 
1216 	res = e820_res;
1217 	for (i = 0; i < e820_table->nr_entries; i++) {
1218 		if (!res->parent && res->end)
1219 			insert_resource_expand_to_fit(&iomem_resource, res);
1220 		res++;
1221 	}
1222 
1223 	/*
1224 	 * Try to bump up RAM regions to reasonable boundaries, to
1225 	 * avoid stolen RAM:
1226 	 */
1227 	for (i = 0; i < e820_table->nr_entries; i++) {
1228 		struct e820_entry *entry = &e820_table->entries[i];
1229 		u64 start, end;
1230 
1231 		if (entry->type != E820_TYPE_RAM)
1232 			continue;
1233 
1234 		start = entry->addr + entry->size;
1235 		end = round_up(start, ram_alignment(start)) - 1;
1236 		if (end > MAX_RESOURCE_SIZE)
1237 			end = MAX_RESOURCE_SIZE;
1238 		if (start >= end)
1239 			continue;
1240 
1241 		printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1242 		reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1243 	}
1244 }
1245 
1246 /*
1247  * Pass the firmware (bootloader) E820 map to the kernel and process it:
1248  */
e820__memory_setup_default(void)1249 char *__init e820__memory_setup_default(void)
1250 {
1251 	char *who = "BIOS-e820";
1252 
1253 	/*
1254 	 * Try to copy the BIOS-supplied E820-map.
1255 	 *
1256 	 * Otherwise fake a memory map; one section from 0k->640k,
1257 	 * the next section from 1mb->appropriate_mem_k
1258 	 */
1259 	if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1260 		u64 mem_size;
1261 
1262 		/* Compare results from other methods and take the one that gives more RAM: */
1263 		if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1264 			mem_size = boot_params.screen_info.ext_mem_k;
1265 			who = "BIOS-88";
1266 		} else {
1267 			mem_size = boot_params.alt_mem_k;
1268 			who = "BIOS-e801";
1269 		}
1270 
1271 		e820_table->nr_entries = 0;
1272 		e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1273 		e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1274 	}
1275 
1276 	/* We just appended a lot of ranges, sanitize the table: */
1277 	e820__update_table(e820_table);
1278 
1279 	return who;
1280 }
1281 
1282 /*
1283  * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1284  * E820 map - with an optional platform quirk available for virtual platforms
1285  * to override this method of boot environment processing:
1286  */
e820__memory_setup(void)1287 void __init e820__memory_setup(void)
1288 {
1289 	char *who;
1290 
1291 	/* This is a firmware interface ABI - make sure we don't break it: */
1292 	BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1293 
1294 	who = x86_init.resources.memory_setup();
1295 
1296 	memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1297 	memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1298 
1299 	pr_info("BIOS-provided physical RAM map:\n");
1300 	e820__print_table(who);
1301 }
1302 
e820__memblock_setup(void)1303 void __init e820__memblock_setup(void)
1304 {
1305 	int i;
1306 	u64 end;
1307 
1308 	/*
1309 	 * The bootstrap memblock region count maximum is 128 entries
1310 	 * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1311 	 * than that - so allow memblock resizing.
1312 	 *
1313 	 * This is safe, because this call happens pretty late during x86 setup,
1314 	 * so we know about reserved memory regions already. (This is important
1315 	 * so that memblock resizing does no stomp over reserved areas.)
1316 	 */
1317 	memblock_allow_resize();
1318 
1319 	for (i = 0; i < e820_table->nr_entries; i++) {
1320 		struct e820_entry *entry = &e820_table->entries[i];
1321 
1322 		end = entry->addr + entry->size;
1323 		if (end != (resource_size_t)end)
1324 			continue;
1325 
1326 		if (entry->type == E820_TYPE_SOFT_RESERVED)
1327 			memblock_reserve(entry->addr, entry->size);
1328 
1329 		if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1330 			continue;
1331 
1332 		memblock_add(entry->addr, entry->size);
1333 	}
1334 
1335 	/* Throw away partial pages: */
1336 	memblock_trim_memory(PAGE_SIZE);
1337 
1338 	memblock_dump_all();
1339 }
1340