2 * Handle the memory map.
3 * The functions here do the job until bootmem takes over.
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>
11 #include <linux/kernel.h>
12 #include <linux/types.h>
13 #include <linux/init.h>
14 #include <linux/bootmem.h>
15 #include <linux/ioport.h>
16 #include <linux/string.h>
17 #include <linux/kexec.h>
18 #include <linux/module.h>
20 #include <linux/pfn.h>
21 #include <linux/suspend.h>
22 #include <linux/firmware-map.h>
24 #include <asm/pgtable.h>
27 #include <asm/proto.h>
28 #include <asm/setup.h>
29 #include <asm/trampoline.h>
32 * The e820 map is the map that gets modified e.g. with command line parameters
33 * and that is also registered with modifications in the kernel resource tree
34 * with the iomem_resource as parent.
36 * The e820_saved is directly saved after the BIOS-provided memory map is
37 * copied. It doesn't get modified afterwards. It's registered for the
38 * /sys/firmware/memmap interface.
40 * That memory map is not modified and is used as base for kexec. The kexec'd
41 * kernel should get the same memory map as the firmware provides. Then the
42 * user can e.g. boot the original kernel with mem=1G while still booting the
43 * next kernel with full memory.
46 struct e820map e820_saved
;
48 /* For PCI or other memory-mapped resources */
49 unsigned long pci_mem_start
= 0xaeedbabe;
51 EXPORT_SYMBOL(pci_mem_start
);
55 * This function checks if any part of the range <start,end> is mapped
59 e820_any_mapped(u64 start
, u64 end
, unsigned type
)
63 for (i
= 0; i
< e820
.nr_map
; i
++) {
64 struct e820entry
*ei
= &e820
.map
[i
];
66 if (type
&& ei
->type
!= type
)
68 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
74 EXPORT_SYMBOL_GPL(e820_any_mapped
);
77 * This function checks if the entire range <start,end> is mapped with type.
79 * Note: this function only works correct if the e820 table is sorted and
80 * not-overlapping, which is the case
82 int __init
e820_all_mapped(u64 start
, u64 end
, unsigned type
)
86 for (i
= 0; i
< e820
.nr_map
; i
++) {
87 struct e820entry
*ei
= &e820
.map
[i
];
89 if (type
&& ei
->type
!= type
)
91 /* is the region (part) in overlap with the current region ?*/
92 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
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
98 if (ei
->addr
<= start
)
99 start
= ei
->addr
+ ei
->size
;
101 * if start is now at or beyond end, we're done, full
111 * Add a memory region to the kernel e820 map.
113 void __init
e820_add_region(u64 start
, u64 size
, int type
)
117 if (x
== ARRAY_SIZE(e820
.map
)) {
118 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
122 e820
.map
[x
].addr
= start
;
123 e820
.map
[x
].size
= size
;
124 e820
.map
[x
].type
= type
;
128 void __init
e820_print_map(char *who
)
132 for (i
= 0; i
< e820
.nr_map
; i
++) {
133 printk(KERN_INFO
" %s: %016Lx - %016Lx ", who
,
134 (unsigned long long) e820
.map
[i
].addr
,
136 (e820
.map
[i
].addr
+ e820
.map
[i
].size
));
137 switch (e820
.map
[i
].type
) {
139 case E820_RESERVED_KERN
:
140 printk(KERN_CONT
"(usable)\n");
143 printk(KERN_CONT
"(reserved)\n");
146 printk(KERN_CONT
"(ACPI data)\n");
149 printk(KERN_CONT
"(ACPI NVS)\n");
152 printk("(unusable)\n");
155 printk(KERN_CONT
"type %u\n", e820
.map
[i
].type
);
162 * Sanitize the BIOS e820 map.
164 * Some e820 responses include overlapping entries. The following
165 * replaces the original e820 map with a new one, removing overlaps,
166 * and resolving conflicting memory types in favor of highest
169 * The input parameter biosmap points to an array of 'struct
170 * e820entry' which on entry has elements in the range [0, *pnr_map)
171 * valid, and which has space for up to max_nr_map entries.
172 * On return, the resulting sanitized e820 map entries will be in
173 * overwritten in the same location, starting at biosmap.
175 * The integer pointed to by pnr_map must be valid on entry (the
176 * current number of valid entries located at biosmap) and will
177 * be updated on return, with the new number of valid entries
178 * (something no more than max_nr_map.)
180 * The return value from sanitize_e820_map() is zero if it
181 * successfully 'sanitized' the map entries passed in, and is -1
182 * if it did nothing, which can happen if either of (1) it was
183 * only passed one map entry, or (2) any of the input map entries
184 * were invalid (start + size < start, meaning that the size was
185 * so big the described memory range wrapped around through zero.)
187 * Visually we're performing the following
188 * (1,2,3,4 = memory types)...
190 * Sample memory map (w/overlaps):
191 * ____22__________________
192 * ______________________4_
193 * ____1111________________
194 * _44_____________________
195 * 11111111________________
196 * ____________________33__
197 * ___________44___________
198 * __________33333_________
199 * ______________22________
200 * ___________________2222_
201 * _________111111111______
202 * _____________________11_
203 * _________________4______
205 * Sanitized equivalent (no overlap):
206 * 1_______________________
207 * _44_____________________
208 * ___1____________________
209 * ____22__________________
210 * ______11________________
211 * _________1______________
212 * __________3_____________
213 * ___________44___________
214 * _____________33_________
215 * _______________2________
216 * ________________1_______
217 * _________________4______
218 * ___________________2____
219 * ____________________33__
220 * ______________________4_
223 int __init
sanitize_e820_map(struct e820entry
*biosmap
, int max_nr_map
,
226 struct change_member
{
227 struct e820entry
*pbios
; /* pointer to original bios entry */
228 unsigned long long addr
; /* address for this change point */
230 static struct change_member change_point_list
[2*E820_X_MAX
] __initdata
;
231 static struct change_member
*change_point
[2*E820_X_MAX
] __initdata
;
232 static struct e820entry
*overlap_list
[E820_X_MAX
] __initdata
;
233 static struct e820entry new_bios
[E820_X_MAX
] __initdata
;
234 struct change_member
*change_tmp
;
235 unsigned long current_type
, last_type
;
236 unsigned long long last_addr
;
237 int chgidx
, still_changing
;
240 int old_nr
, new_nr
, chg_nr
;
243 /* if there's only one memory region, don't bother */
248 BUG_ON(old_nr
> max_nr_map
);
250 /* bail out if we find any unreasonable addresses in bios map */
251 for (i
= 0; i
< old_nr
; i
++)
252 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
)
255 /* create pointers for initial change-point information (for sorting) */
256 for (i
= 0; i
< 2 * old_nr
; i
++)
257 change_point
[i
] = &change_point_list
[i
];
259 /* record all known change-points (starting and ending addresses),
260 omitting those that are for empty memory regions */
262 for (i
= 0; i
< old_nr
; i
++) {
263 if (biosmap
[i
].size
!= 0) {
264 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
265 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
266 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+
268 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
273 /* sort change-point list by memory addresses (low -> high) */
275 while (still_changing
) {
277 for (i
= 1; i
< chg_nr
; i
++) {
278 unsigned long long curaddr
, lastaddr
;
279 unsigned long long curpbaddr
, lastpbaddr
;
281 curaddr
= change_point
[i
]->addr
;
282 lastaddr
= change_point
[i
- 1]->addr
;
283 curpbaddr
= change_point
[i
]->pbios
->addr
;
284 lastpbaddr
= change_point
[i
- 1]->pbios
->addr
;
287 * swap entries, when:
289 * curaddr > lastaddr or
290 * curaddr == lastaddr and curaddr == curpbaddr and
291 * lastaddr != lastpbaddr
293 if (curaddr
< lastaddr
||
294 (curaddr
== lastaddr
&& curaddr
== curpbaddr
&&
295 lastaddr
!= lastpbaddr
)) {
296 change_tmp
= change_point
[i
];
297 change_point
[i
] = change_point
[i
-1];
298 change_point
[i
-1] = change_tmp
;
304 /* create a new bios memory map, removing overlaps */
305 overlap_entries
= 0; /* number of entries in the overlap table */
306 new_bios_entry
= 0; /* index for creating new bios map entries */
307 last_type
= 0; /* start with undefined memory type */
308 last_addr
= 0; /* start with 0 as last starting address */
310 /* loop through change-points, determining affect on the new bios map */
311 for (chgidx
= 0; chgidx
< chg_nr
; chgidx
++) {
312 /* keep track of all overlapping bios entries */
313 if (change_point
[chgidx
]->addr
==
314 change_point
[chgidx
]->pbios
->addr
) {
316 * add map entry to overlap list (> 1 entry
317 * implies an overlap)
319 overlap_list
[overlap_entries
++] =
320 change_point
[chgidx
]->pbios
;
323 * remove entry from list (order independent,
326 for (i
= 0; i
< overlap_entries
; i
++) {
327 if (overlap_list
[i
] ==
328 change_point
[chgidx
]->pbios
)
330 overlap_list
[overlap_entries
-1];
335 * if there are overlapping entries, decide which
336 * "type" to use (larger value takes precedence --
337 * 1=usable, 2,3,4,4+=unusable)
340 for (i
= 0; i
< overlap_entries
; i
++)
341 if (overlap_list
[i
]->type
> current_type
)
342 current_type
= overlap_list
[i
]->type
;
344 * continue building up new bios map based on this
347 if (current_type
!= last_type
) {
348 if (last_type
!= 0) {
349 new_bios
[new_bios_entry
].size
=
350 change_point
[chgidx
]->addr
- last_addr
;
352 * move forward only if the new size
355 if (new_bios
[new_bios_entry
].size
!= 0)
357 * no more space left for new
360 if (++new_bios_entry
>= max_nr_map
)
363 if (current_type
!= 0) {
364 new_bios
[new_bios_entry
].addr
=
365 change_point
[chgidx
]->addr
;
366 new_bios
[new_bios_entry
].type
= current_type
;
367 last_addr
= change_point
[chgidx
]->addr
;
369 last_type
= current_type
;
372 /* retain count for new bios entries */
373 new_nr
= new_bios_entry
;
375 /* copy new bios mapping into original location */
376 memcpy(biosmap
, new_bios
, new_nr
* sizeof(struct e820entry
));
382 static int __init
__append_e820_map(struct e820entry
*biosmap
, int nr_map
)
385 u64 start
= biosmap
->addr
;
386 u64 size
= biosmap
->size
;
387 u64 end
= start
+ size
;
388 u32 type
= biosmap
->type
;
390 /* Overflow in 64 bits? Ignore the memory map. */
394 e820_add_region(start
, size
, type
);
403 * Copy the BIOS e820 map into a safe place.
405 * Sanity-check it while we're at it..
407 * If we're lucky and live on a modern system, the setup code
408 * will have given us a memory map that we can use to properly
409 * set up memory. If we aren't, we'll fake a memory map.
411 static int __init
append_e820_map(struct e820entry
*biosmap
, int nr_map
)
413 /* Only one memory region (or negative)? Ignore it */
417 return __append_e820_map(biosmap
, nr_map
);
420 static u64 __init
e820_update_range_map(struct e820map
*e820x
, u64 start
,
421 u64 size
, unsigned old_type
,
425 u64 real_updated_size
= 0;
427 BUG_ON(old_type
== new_type
);
429 if (size
> (ULLONG_MAX
- start
))
430 size
= ULLONG_MAX
- start
;
432 for (i
= 0; i
< e820x
->nr_map
; i
++) {
433 struct e820entry
*ei
= &e820x
->map
[i
];
434 u64 final_start
, final_end
;
435 if (ei
->type
!= old_type
)
437 /* totally covered? */
438 if (ei
->addr
>= start
&&
439 (ei
->addr
+ ei
->size
) <= (start
+ size
)) {
441 real_updated_size
+= ei
->size
;
444 /* partially covered */
445 final_start
= max(start
, ei
->addr
);
446 final_end
= min(start
+ size
, ei
->addr
+ ei
->size
);
447 if (final_start
>= final_end
)
451 if (x
== ARRAY_SIZE(e820x
->map
)) {
452 printk(KERN_ERR
"Too many memory map entries!\n");
455 e820x
->map
[x
].addr
= final_start
;
456 e820x
->map
[x
].size
= final_end
- final_start
;
457 e820x
->map
[x
].type
= new_type
;
460 real_updated_size
+= final_end
- final_start
;
462 if (ei
->addr
< final_start
)
464 ei
->addr
= final_end
;
465 ei
->size
-= final_end
- final_start
;
467 return real_updated_size
;
470 u64 __init
e820_update_range(u64 start
, u64 size
, unsigned old_type
,
473 return e820_update_range_map(&e820
, start
, size
, old_type
, new_type
);
476 static u64 __init
e820_update_range_saved(u64 start
, u64 size
,
477 unsigned old_type
, unsigned new_type
)
479 return e820_update_range_map(&e820_saved
, start
, size
, old_type
,
483 /* make e820 not cover the range */
484 u64 __init
e820_remove_range(u64 start
, u64 size
, unsigned old_type
,
488 u64 real_removed_size
= 0;
490 if (size
> (ULLONG_MAX
- start
))
491 size
= ULLONG_MAX
- start
;
493 for (i
= 0; i
< e820
.nr_map
; i
++) {
494 struct e820entry
*ei
= &e820
.map
[i
];
495 u64 final_start
, final_end
;
497 if (checktype
&& ei
->type
!= old_type
)
499 /* totally covered? */
500 if (ei
->addr
>= start
&&
501 (ei
->addr
+ ei
->size
) <= (start
+ size
)) {
502 real_removed_size
+= ei
->size
;
503 memset(ei
, 0, sizeof(struct e820entry
));
506 /* partially covered */
507 final_start
= max(start
, ei
->addr
);
508 final_end
= min(start
+ size
, ei
->addr
+ ei
->size
);
509 if (final_start
>= final_end
)
511 real_removed_size
+= final_end
- final_start
;
513 ei
->size
-= final_end
- final_start
;
514 if (ei
->addr
< final_start
)
516 ei
->addr
= final_end
;
518 return real_removed_size
;
521 void __init
update_e820(void)
525 nr_map
= e820
.nr_map
;
526 if (sanitize_e820_map(e820
.map
, ARRAY_SIZE(e820
.map
), &nr_map
))
528 e820
.nr_map
= nr_map
;
529 printk(KERN_INFO
"modified physical RAM map:\n");
530 e820_print_map("modified");
532 static void __init
update_e820_saved(void)
536 nr_map
= e820_saved
.nr_map
;
537 if (sanitize_e820_map(e820_saved
.map
, ARRAY_SIZE(e820_saved
.map
), &nr_map
))
539 e820_saved
.nr_map
= nr_map
;
541 #define MAX_GAP_END 0x100000000ull
543 * Search for a gap in the e820 memory space from start_addr to end_addr.
545 __init
int e820_search_gap(unsigned long *gapstart
, unsigned long *gapsize
,
546 unsigned long start_addr
, unsigned long long end_addr
)
548 unsigned long long last
;
552 last
= (end_addr
&& end_addr
< MAX_GAP_END
) ? end_addr
: MAX_GAP_END
;
555 unsigned long long start
= e820
.map
[i
].addr
;
556 unsigned long long end
= start
+ e820
.map
[i
].size
;
558 if (end
< start_addr
)
562 * Since "last" is at most 4GB, we know we'll
563 * fit in 32 bits if this condition is true
566 unsigned long gap
= last
- end
;
568 if (gap
>= *gapsize
) {
581 * Search for the biggest gap in the low 32 bits of the e820
582 * memory space. We pass this space to PCI to assign MMIO resources
583 * for hotplug or unconfigured devices in.
584 * Hopefully the BIOS let enough space left.
586 __init
void e820_setup_gap(void)
588 unsigned long gapstart
, gapsize
, round
;
591 gapstart
= 0x10000000;
593 found
= e820_search_gap(&gapstart
, &gapsize
, 0, MAX_GAP_END
);
597 gapstart
= (max_pfn
<< PAGE_SHIFT
) + 1024*1024;
598 printk(KERN_ERR
"PCI: Warning: Cannot find a gap in the 32bit "
600 KERN_ERR
"PCI: Unassigned devices with 32bit resource "
601 "registers may break!\n");
606 * See how much we want to round up: start off with
607 * rounding to the next 1MB area.
610 while ((gapsize
>> 4) > round
)
612 /* Fun with two's complement */
613 pci_mem_start
= (gapstart
+ round
) & -round
;
616 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
617 pci_mem_start
, gapstart
, gapsize
);
621 * Because of the size limitation of struct boot_params, only first
622 * 128 E820 memory entries are passed to kernel via
623 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
624 * linked list of struct setup_data, which is parsed here.
626 void __init
parse_e820_ext(struct setup_data
*sdata
, unsigned long pa_data
)
630 struct e820entry
*extmap
;
632 entries
= sdata
->len
/ sizeof(struct e820entry
);
633 map_len
= sdata
->len
+ sizeof(struct setup_data
);
634 if (map_len
> PAGE_SIZE
)
635 sdata
= early_ioremap(pa_data
, map_len
);
636 extmap
= (struct e820entry
*)(sdata
->data
);
637 __append_e820_map(extmap
, entries
);
638 sanitize_e820_map(e820
.map
, ARRAY_SIZE(e820
.map
), &e820
.nr_map
);
639 if (map_len
> PAGE_SIZE
)
640 early_iounmap(sdata
, map_len
);
641 printk(KERN_INFO
"extended physical RAM map:\n");
642 e820_print_map("extended");
645 #if defined(CONFIG_X86_64) || \
646 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
648 * Find the ranges of physical addresses that do not correspond to
649 * e820 RAM areas and mark the corresponding pages as nosave for
650 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
652 * This function requires the e820 map to be sorted and without any
653 * overlapping entries and assumes the first e820 area to be RAM.
655 void __init
e820_mark_nosave_regions(unsigned long limit_pfn
)
660 pfn
= PFN_DOWN(e820
.map
[0].addr
+ e820
.map
[0].size
);
661 for (i
= 1; i
< e820
.nr_map
; i
++) {
662 struct e820entry
*ei
= &e820
.map
[i
];
664 if (pfn
< PFN_UP(ei
->addr
))
665 register_nosave_region(pfn
, PFN_UP(ei
->addr
));
667 pfn
= PFN_DOWN(ei
->addr
+ ei
->size
);
668 if (ei
->type
!= E820_RAM
&& ei
->type
!= E820_RESERVED_KERN
)
669 register_nosave_region(PFN_UP(ei
->addr
), pfn
);
671 if (pfn
>= limit_pfn
)
677 #ifdef CONFIG_HIBERNATION
679 * Mark ACPI NVS memory region, so that we can save/restore it during
680 * hibernation and the subsequent resume.
682 static int __init
e820_mark_nvs_memory(void)
686 for (i
= 0; i
< e820
.nr_map
; i
++) {
687 struct e820entry
*ei
= &e820
.map
[i
];
689 if (ei
->type
== E820_NVS
)
690 hibernate_nvs_register(ei
->addr
, ei
->size
);
695 core_initcall(e820_mark_nvs_memory
);
699 * Early reserved memory areas.
701 #define MAX_EARLY_RES 20
708 static struct early_res early_res
[MAX_EARLY_RES
] __initdata
= {
709 { 0, PAGE_SIZE
, "BIOS data page" }, /* BIOS data page */
713 static int __init
find_overlapped_early(u64 start
, u64 end
)
718 for (i
= 0; i
< MAX_EARLY_RES
&& early_res
[i
].end
; i
++) {
720 if (end
> r
->start
&& start
< r
->end
)
728 * Drop the i-th range from the early reservation map,
729 * by copying any higher ranges down one over it, and
730 * clearing what had been the last slot.
732 static void __init
drop_range(int i
)
736 for (j
= i
+ 1; j
< MAX_EARLY_RES
&& early_res
[j
].end
; j
++)
739 memmove(&early_res
[i
], &early_res
[i
+ 1],
740 (j
- 1 - i
) * sizeof(struct early_res
));
742 early_res
[j
- 1].end
= 0;
746 * Split any existing ranges that:
747 * 1) are marked 'overlap_ok', and
748 * 2) overlap with the stated range [start, end)
749 * into whatever portion (if any) of the existing range is entirely
750 * below or entirely above the stated range. Drop the portion
751 * of the existing range that overlaps with the stated range,
752 * which will allow the caller of this routine to then add that
753 * stated range without conflicting with any existing range.
755 static void __init
drop_overlaps_that_are_ok(u64 start
, u64 end
)
759 u64 lower_start
, lower_end
;
760 u64 upper_start
, upper_end
;
763 for (i
= 0; i
< MAX_EARLY_RES
&& early_res
[i
].end
; i
++) {
766 /* Continue past non-overlapping ranges */
767 if (end
<= r
->start
|| start
>= r
->end
)
771 * Leave non-ok overlaps as is; let caller
772 * panic "Overlapping early reservations"
773 * when it hits this overlap.
779 * We have an ok overlap. We will drop it from the early
780 * reservation map, and add back in any non-overlapping
781 * portions (lower or upper) as separate, overlap_ok,
782 * non-overlapping ranges.
785 /* 1. Note any non-overlapping (lower or upper) ranges. */
786 strncpy(name
, r
->name
, sizeof(name
) - 1);
788 lower_start
= lower_end
= 0;
789 upper_start
= upper_end
= 0;
790 if (r
->start
< start
) {
791 lower_start
= r
->start
;
799 /* 2. Drop the original ok overlapping range */
802 i
--; /* resume for-loop on copied down entry */
804 /* 3. Add back in any non-overlapping ranges. */
806 reserve_early_overlap_ok(lower_start
, lower_end
, name
);
808 reserve_early_overlap_ok(upper_start
, upper_end
, name
);
812 static void __init
__reserve_early(u64 start
, u64 end
, char *name
,
818 i
= find_overlapped_early(start
, end
);
819 if (i
>= MAX_EARLY_RES
)
820 panic("Too many early reservations");
823 panic("Overlapping early reservations "
824 "%llx-%llx %s to %llx-%llx %s\n",
825 start
, end
- 1, name
?name
:"", r
->start
,
826 r
->end
- 1, r
->name
);
829 r
->overlap_ok
= overlap_ok
;
831 strncpy(r
->name
, name
, sizeof(r
->name
) - 1);
835 * A few early reservtations come here.
837 * The 'overlap_ok' in the name of this routine does -not- mean it
838 * is ok for these reservations to overlap an earlier reservation.
839 * Rather it means that it is ok for subsequent reservations to
842 * Use this entry point to reserve early ranges when you are doing
843 * so out of "Paranoia", reserving perhaps more memory than you need,
844 * just in case, and don't mind a subsequent overlapping reservation
845 * that is known to be needed.
847 * The drop_overlaps_that_are_ok() call here isn't really needed.
848 * It would be needed if we had two colliding 'overlap_ok'
849 * reservations, so that the second such would not panic on the
850 * overlap with the first. We don't have any such as of this
851 * writing, but might as well tolerate such if it happens in
854 void __init
reserve_early_overlap_ok(u64 start
, u64 end
, char *name
)
856 drop_overlaps_that_are_ok(start
, end
);
857 __reserve_early(start
, end
, name
, 1);
861 * Most early reservations come here.
863 * We first have drop_overlaps_that_are_ok() drop any pre-existing
864 * 'overlap_ok' ranges, so that we can then reserve this memory
865 * range without risk of panic'ing on an overlapping overlap_ok
868 void __init
reserve_early(u64 start
, u64 end
, char *name
)
873 drop_overlaps_that_are_ok(start
, end
);
874 __reserve_early(start
, end
, name
, 0);
877 void __init
free_early(u64 start
, u64 end
)
882 i
= find_overlapped_early(start
, end
);
884 if (i
>= MAX_EARLY_RES
|| r
->end
!= end
|| r
->start
!= start
)
885 panic("free_early on not reserved area: %llx-%llx!",
891 void __init
early_res_to_bootmem(u64 start
, u64 end
)
894 u64 final_start
, final_end
;
897 for (i
= 0; i
< MAX_EARLY_RES
&& early_res
[i
].end
; i
++)
900 printk(KERN_INFO
"(%d early reservations) ==> bootmem [%010llx - %010llx]\n",
902 for (i
= 0; i
< count
; i
++) {
903 struct early_res
*r
= &early_res
[i
];
904 printk(KERN_INFO
" #%d [%010llx - %010llx] %16s", i
,
905 r
->start
, r
->end
, r
->name
);
906 final_start
= max(start
, r
->start
);
907 final_end
= min(end
, r
->end
);
908 if (final_start
>= final_end
) {
909 printk(KERN_CONT
"\n");
912 printk(KERN_CONT
" ==> [%010llx - %010llx]\n",
913 final_start
, final_end
);
914 reserve_bootmem_generic(final_start
, final_end
- final_start
,
919 /* Check for already reserved areas */
920 static inline int __init
bad_addr(u64
*addrp
, u64 size
, u64 align
)
927 i
= find_overlapped_early(addr
, addr
+ size
);
929 if (i
< MAX_EARLY_RES
&& r
->end
) {
930 *addrp
= addr
= round_up(r
->end
, align
);
937 /* Check for already reserved areas */
938 static inline int __init
bad_addr_size(u64
*addrp
, u64
*sizep
, u64 align
)
941 u64 addr
= *addrp
, last
;
946 for (i
= 0; i
< MAX_EARLY_RES
&& early_res
[i
].end
; i
++) {
947 struct early_res
*r
= &early_res
[i
];
948 if (last
> r
->start
&& addr
< r
->start
) {
949 size
= r
->start
- addr
;
953 if (last
> r
->end
&& addr
< r
->end
) {
954 addr
= round_up(r
->end
, align
);
959 if (last
<= r
->end
&& addr
>= r
->start
) {
972 * Find a free area with specified alignment in a specific range.
974 u64 __init
find_e820_area(u64 start
, u64 end
, u64 size
, u64 align
)
978 for (i
= 0; i
< e820
.nr_map
; i
++) {
979 struct e820entry
*ei
= &e820
.map
[i
];
983 if (ei
->type
!= E820_RAM
)
985 addr
= round_up(ei
->addr
, align
);
986 ei_last
= ei
->addr
+ ei
->size
;
988 addr
= round_up(start
, align
);
991 while (bad_addr(&addr
, size
, align
) && addr
+size
<= ei_last
)
1004 * Find next free range after *start
1006 u64 __init
find_e820_area_size(u64 start
, u64
*sizep
, u64 align
)
1010 for (i
= 0; i
< e820
.nr_map
; i
++) {
1011 struct e820entry
*ei
= &e820
.map
[i
];
1015 if (ei
->type
!= E820_RAM
)
1017 addr
= round_up(ei
->addr
, align
);
1018 ei_last
= ei
->addr
+ ei
->size
;
1020 addr
= round_up(start
, align
);
1021 if (addr
>= ei_last
)
1023 *sizep
= ei_last
- addr
;
1024 while (bad_addr_size(&addr
, sizep
, align
) &&
1025 addr
+ *sizep
<= ei_last
)
1027 last
= addr
+ *sizep
;
1037 * pre allocated 4k and reserved it in e820
1039 u64 __init
early_reserve_e820(u64 startt
, u64 sizet
, u64 align
)
1046 while (size
< sizet
&& (start
+ 1))
1047 start
= find_e820_area_size(start
, &size
, align
);
1052 #ifdef CONFIG_X86_32
1053 if (start
>= MAXMEM
)
1055 if (start
+ size
> MAXMEM
)
1056 size
= MAXMEM
- start
;
1059 addr
= round_down(start
+ size
- sizet
, align
);
1062 e820_update_range(addr
, sizet
, E820_RAM
, E820_RESERVED
);
1063 e820_update_range_saved(addr
, sizet
, E820_RAM
, E820_RESERVED
);
1064 printk(KERN_INFO
"update e820 for early_reserve_e820\n");
1066 update_e820_saved();
1071 #ifdef CONFIG_X86_32
1072 # ifdef CONFIG_X86_PAE
1073 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
1075 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
1077 #else /* CONFIG_X86_32 */
1078 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
1082 * Find the highest page frame number we have available
1084 static unsigned long __init
e820_end_pfn(unsigned long limit_pfn
, unsigned type
)
1087 unsigned long last_pfn
= 0;
1088 unsigned long max_arch_pfn
= MAX_ARCH_PFN
;
1090 for (i
= 0; i
< e820
.nr_map
; i
++) {
1091 struct e820entry
*ei
= &e820
.map
[i
];
1092 unsigned long start_pfn
;
1093 unsigned long end_pfn
;
1095 if (ei
->type
!= type
)
1098 start_pfn
= ei
->addr
>> PAGE_SHIFT
;
1099 end_pfn
= (ei
->addr
+ ei
->size
) >> PAGE_SHIFT
;
1101 if (start_pfn
>= limit_pfn
)
1103 if (end_pfn
> limit_pfn
) {
1104 last_pfn
= limit_pfn
;
1107 if (end_pfn
> last_pfn
)
1111 if (last_pfn
> max_arch_pfn
)
1112 last_pfn
= max_arch_pfn
;
1114 printk(KERN_INFO
"last_pfn = %#lx max_arch_pfn = %#lx\n",
1115 last_pfn
, max_arch_pfn
);
1118 unsigned long __init
e820_end_of_ram_pfn(void)
1120 return e820_end_pfn(MAX_ARCH_PFN
, E820_RAM
);
1123 unsigned long __init
e820_end_of_low_ram_pfn(void)
1125 return e820_end_pfn(1UL<<(32 - PAGE_SHIFT
), E820_RAM
);
1128 * Finds an active region in the address range from start_pfn to last_pfn and
1129 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
1131 int __init
e820_find_active_region(const struct e820entry
*ei
,
1132 unsigned long start_pfn
,
1133 unsigned long last_pfn
,
1134 unsigned long *ei_startpfn
,
1135 unsigned long *ei_endpfn
)
1137 u64 align
= PAGE_SIZE
;
1139 *ei_startpfn
= round_up(ei
->addr
, align
) >> PAGE_SHIFT
;
1140 *ei_endpfn
= round_down(ei
->addr
+ ei
->size
, align
) >> PAGE_SHIFT
;
1142 /* Skip map entries smaller than a page */
1143 if (*ei_startpfn
>= *ei_endpfn
)
1146 /* Skip if map is outside the node */
1147 if (ei
->type
!= E820_RAM
|| *ei_endpfn
<= start_pfn
||
1148 *ei_startpfn
>= last_pfn
)
1151 /* Check for overlaps */
1152 if (*ei_startpfn
< start_pfn
)
1153 *ei_startpfn
= start_pfn
;
1154 if (*ei_endpfn
> last_pfn
)
1155 *ei_endpfn
= last_pfn
;
1160 /* Walk the e820 map and register active regions within a node */
1161 void __init
e820_register_active_regions(int nid
, unsigned long start_pfn
,
1162 unsigned long last_pfn
)
1164 unsigned long ei_startpfn
;
1165 unsigned long ei_endpfn
;
1168 for (i
= 0; i
< e820
.nr_map
; i
++)
1169 if (e820_find_active_region(&e820
.map
[i
],
1170 start_pfn
, last_pfn
,
1171 &ei_startpfn
, &ei_endpfn
))
1172 add_active_range(nid
, ei_startpfn
, ei_endpfn
);
1176 * Find the hole size (in bytes) in the memory range.
1177 * @start: starting address of the memory range to scan
1178 * @end: ending address of the memory range to scan
1180 u64 __init
e820_hole_size(u64 start
, u64 end
)
1182 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
1183 unsigned long last_pfn
= end
>> PAGE_SHIFT
;
1184 unsigned long ei_startpfn
, ei_endpfn
, ram
= 0;
1187 for (i
= 0; i
< e820
.nr_map
; i
++) {
1188 if (e820_find_active_region(&e820
.map
[i
],
1189 start_pfn
, last_pfn
,
1190 &ei_startpfn
, &ei_endpfn
))
1191 ram
+= ei_endpfn
- ei_startpfn
;
1193 return end
- start
- ((u64
)ram
<< PAGE_SHIFT
);
1196 static void early_panic(char *msg
)
1202 static int userdef __initdata
;
1204 /* "mem=nopentium" disables the 4MB page tables. */
1205 static int __init
parse_memopt(char *p
)
1212 #ifdef CONFIG_X86_32
1213 if (!strcmp(p
, "nopentium")) {
1214 setup_clear_cpu_cap(X86_FEATURE_PSE
);
1220 mem_size
= memparse(p
, &p
);
1221 e820_remove_range(mem_size
, ULLONG_MAX
- mem_size
, E820_RAM
, 1);
1225 early_param("mem", parse_memopt
);
1227 static int __init
parse_memmap_opt(char *p
)
1230 u64 start_at
, mem_size
;
1235 if (!strncmp(p
, "exactmap", 8)) {
1236 #ifdef CONFIG_CRASH_DUMP
1238 * If we are doing a crash dump, we still need to know
1239 * the real mem size before original memory map is
1242 saved_max_pfn
= e820_end_of_ram_pfn();
1250 mem_size
= memparse(p
, &p
);
1256 start_at
= memparse(p
+1, &p
);
1257 e820_add_region(start_at
, mem_size
, E820_RAM
);
1258 } else if (*p
== '#') {
1259 start_at
= memparse(p
+1, &p
);
1260 e820_add_region(start_at
, mem_size
, E820_ACPI
);
1261 } else if (*p
== '$') {
1262 start_at
= memparse(p
+1, &p
);
1263 e820_add_region(start_at
, mem_size
, E820_RESERVED
);
1265 e820_remove_range(mem_size
, ULLONG_MAX
- mem_size
, E820_RAM
, 1);
1267 return *p
== '\0' ? 0 : -EINVAL
;
1269 early_param("memmap", parse_memmap_opt
);
1271 void __init
finish_e820_parsing(void)
1274 int nr
= e820
.nr_map
;
1276 if (sanitize_e820_map(e820
.map
, ARRAY_SIZE(e820
.map
), &nr
) < 0)
1277 early_panic("Invalid user supplied memory map");
1280 printk(KERN_INFO
"user-defined physical RAM map:\n");
1281 e820_print_map("user");
1285 static inline const char *e820_type_to_string(int e820_type
)
1287 switch (e820_type
) {
1288 case E820_RESERVED_KERN
:
1289 case E820_RAM
: return "System RAM";
1290 case E820_ACPI
: return "ACPI Tables";
1291 case E820_NVS
: return "ACPI Non-volatile Storage";
1292 case E820_UNUSABLE
: return "Unusable memory";
1293 default: return "reserved";
1298 * Mark e820 reserved areas as busy for the resource manager.
1300 static struct resource __initdata
*e820_res
;
1301 void __init
e820_reserve_resources(void)
1304 struct resource
*res
;
1307 res
= alloc_bootmem_low(sizeof(struct resource
) * e820
.nr_map
);
1309 for (i
= 0; i
< e820
.nr_map
; i
++) {
1310 end
= e820
.map
[i
].addr
+ e820
.map
[i
].size
- 1;
1311 if (end
!= (resource_size_t
)end
) {
1315 res
->name
= e820_type_to_string(e820
.map
[i
].type
);
1316 res
->start
= e820
.map
[i
].addr
;
1319 res
->flags
= IORESOURCE_MEM
;
1322 * don't register the region that could be conflicted with
1323 * pci device BAR resource and insert them later in
1324 * pcibios_resource_survey()
1326 if (e820
.map
[i
].type
!= E820_RESERVED
|| res
->start
< (1ULL<<20)) {
1327 res
->flags
|= IORESOURCE_BUSY
;
1328 insert_resource(&iomem_resource
, res
);
1333 for (i
= 0; i
< e820_saved
.nr_map
; i
++) {
1334 struct e820entry
*entry
= &e820_saved
.map
[i
];
1335 firmware_map_add_early(entry
->addr
,
1336 entry
->addr
+ entry
->size
- 1,
1337 e820_type_to_string(entry
->type
));
1341 void __init
e820_reserve_resources_late(void)
1344 struct resource
*res
;
1347 for (i
= 0; i
< e820
.nr_map
; i
++) {
1348 if (!res
->parent
&& res
->end
)
1349 insert_resource_expand_to_fit(&iomem_resource
, res
);
1354 char *__init
default_machine_specific_memory_setup(void)
1356 char *who
= "BIOS-e820";
1359 * Try to copy the BIOS-supplied E820-map.
1361 * Otherwise fake a memory map; one section from 0k->640k,
1362 * the next section from 1mb->appropriate_mem_k
1364 new_nr
= boot_params
.e820_entries
;
1365 sanitize_e820_map(boot_params
.e820_map
,
1366 ARRAY_SIZE(boot_params
.e820_map
),
1368 boot_params
.e820_entries
= new_nr
;
1369 if (append_e820_map(boot_params
.e820_map
, boot_params
.e820_entries
)
1373 /* compare results from other methods and take the greater */
1374 if (boot_params
.alt_mem_k
1375 < boot_params
.screen_info
.ext_mem_k
) {
1376 mem_size
= boot_params
.screen_info
.ext_mem_k
;
1379 mem_size
= boot_params
.alt_mem_k
;
1384 e820_add_region(0, LOWMEMSIZE(), E820_RAM
);
1385 e820_add_region(HIGH_MEMORY
, mem_size
<< 10, E820_RAM
);
1388 /* In case someone cares... */
1392 char *__init
__attribute__((weak
)) machine_specific_memory_setup(void)
1394 if (x86_quirks
->arch_memory_setup
) {
1395 char *who
= x86_quirks
->arch_memory_setup();
1400 return default_machine_specific_memory_setup();
1403 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
1404 char * __init
__attribute__((weak
)) memory_setup(void)
1406 return machine_specific_memory_setup();
1409 void __init
setup_memory_map(void)
1413 who
= memory_setup();
1414 memcpy(&e820_saved
, &e820
, sizeof(struct e820map
));
1415 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
1416 e820_print_map(who
);