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/suspend.h>
21 #include <linux/pfn.h>
23 #include <asm/pgtable.h>
26 #include <asm/proto.h>
27 #include <asm/setup.h>
28 #include <asm/sections.h>
29 #include <asm/kdebug.h>
34 * PFN of last memory page.
36 unsigned long end_pfn
;
39 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
40 * The direct mapping extends to end_pfn_map, so that we can directly access
41 * apertures, ACPI and other tables without having to play with fixmaps.
43 unsigned long end_pfn_map
;
46 * Last pfn which the user wants to use.
48 static unsigned long __initdata end_user_pfn
= MAXMEM
>>PAGE_SHIFT
;
51 * Early reserved memory areas.
53 #define MAX_EARLY_RES 20
56 unsigned long start
, end
;
58 static struct early_res early_res
[MAX_EARLY_RES
] __initdata
= {
59 { 0, PAGE_SIZE
}, /* BIOS data page */
61 { SMP_TRAMPOLINE_BASE
, SMP_TRAMPOLINE_BASE
+ 2*PAGE_SIZE
},
66 void __init
reserve_early(unsigned long start
, unsigned long end
)
70 for (i
= 0; i
< MAX_EARLY_RES
&& early_res
[i
].end
; i
++) {
72 if (end
> r
->start
&& start
< r
->end
)
73 panic("Duplicated early reservation %lx-%lx\n",
76 if (i
>= MAX_EARLY_RES
)
77 panic("Too many early reservations");
83 void __init
early_res_to_bootmem(void)
86 for (i
= 0; i
< MAX_EARLY_RES
&& early_res
[i
].end
; i
++) {
87 struct early_res
*r
= &early_res
[i
];
88 reserve_bootmem_generic(r
->start
, r
->end
- r
->start
);
92 /* Check for already reserved areas */
93 static inline int bad_addr(unsigned long *addrp
, unsigned long size
)
96 unsigned long addr
= *addrp
, last
;
100 for (i
= 0; i
< MAX_EARLY_RES
&& early_res
[i
].end
; i
++) {
101 struct early_res
*r
= &early_res
[i
];
102 if (last
>= r
->start
&& addr
< r
->end
) {
103 *addrp
= addr
= r
->end
;
112 * This function checks if any part of the range <start,end> is mapped
116 e820_any_mapped(unsigned long start
, unsigned long end
, unsigned type
)
120 for (i
= 0; i
< e820
.nr_map
; i
++) {
121 struct e820entry
*ei
= &e820
.map
[i
];
123 if (type
&& ei
->type
!= type
)
125 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
131 EXPORT_SYMBOL_GPL(e820_any_mapped
);
134 * This function checks if the entire range <start,end> is mapped with type.
136 * Note: this function only works correct if the e820 table is sorted and
137 * not-overlapping, which is the case
139 int __init
e820_all_mapped(unsigned long start
, unsigned long end
,
144 for (i
= 0; i
< e820
.nr_map
; i
++) {
145 struct e820entry
*ei
= &e820
.map
[i
];
147 if (type
&& ei
->type
!= type
)
149 /* is the region (part) in overlap with the current region ?*/
150 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
153 /* if the region is at the beginning of <start,end> we move
154 * start to the end of the region since it's ok until there
156 if (ei
->addr
<= start
)
157 start
= ei
->addr
+ ei
->size
;
159 * if start is now at or beyond end, we're done, full
169 * Find a free area in a specific range.
171 unsigned long __init
find_e820_area(unsigned long start
, unsigned long end
,
176 for (i
= 0; i
< e820
.nr_map
; i
++) {
177 struct e820entry
*ei
= &e820
.map
[i
];
178 unsigned long addr
= ei
->addr
, last
;
180 if (ei
->type
!= E820_RAM
)
184 if (addr
> ei
->addr
+ ei
->size
)
186 while (bad_addr(&addr
, size
) && addr
+size
<= ei
->addr
+ei
->size
)
188 last
= PAGE_ALIGN(addr
) + size
;
189 if (last
> ei
->addr
+ ei
->size
)
199 * Find the highest page frame number we have available
201 unsigned long __init
e820_end_of_ram(void)
203 unsigned long end_pfn
;
205 end_pfn
= find_max_pfn_with_active_regions();
207 if (end_pfn
> end_pfn_map
)
208 end_pfn_map
= end_pfn
;
209 if (end_pfn_map
> MAXMEM
>>PAGE_SHIFT
)
210 end_pfn_map
= MAXMEM
>>PAGE_SHIFT
;
211 if (end_pfn
> end_user_pfn
)
212 end_pfn
= end_user_pfn
;
213 if (end_pfn
> end_pfn_map
)
214 end_pfn
= end_pfn_map
;
216 printk(KERN_INFO
"end_pfn_map = %lu\n", end_pfn_map
);
221 * Mark e820 reserved areas as busy for the resource manager.
223 void __init
e820_reserve_resources(struct resource
*code_resource
,
224 struct resource
*data_resource
, struct resource
*bss_resource
)
227 for (i
= 0; i
< e820
.nr_map
; i
++) {
228 struct resource
*res
;
229 res
= alloc_bootmem_low(sizeof(struct resource
));
230 switch (e820
.map
[i
].type
) {
231 case E820_RAM
: res
->name
= "System RAM"; break;
232 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
233 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
234 default: res
->name
= "reserved";
236 res
->start
= e820
.map
[i
].addr
;
237 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
238 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
239 request_resource(&iomem_resource
, res
);
240 if (e820
.map
[i
].type
== E820_RAM
) {
242 * We don't know which RAM region contains kernel data,
243 * so we try it repeatedly and let the resource manager
246 request_resource(res
, code_resource
);
247 request_resource(res
, data_resource
);
248 request_resource(res
, bss_resource
);
250 if (crashk_res
.start
!= crashk_res
.end
)
251 request_resource(res
, &crashk_res
);
258 * Find the ranges of physical addresses that do not correspond to
259 * e820 RAM areas and mark the corresponding pages as nosave for software
260 * suspend and suspend to RAM.
262 * This function requires the e820 map to be sorted and without any
263 * overlapping entries and assumes the first e820 area to be RAM.
265 void __init
e820_mark_nosave_regions(void)
270 paddr
= round_down(e820
.map
[0].addr
+ e820
.map
[0].size
, PAGE_SIZE
);
271 for (i
= 1; i
< e820
.nr_map
; i
++) {
272 struct e820entry
*ei
= &e820
.map
[i
];
274 if (paddr
< ei
->addr
)
275 register_nosave_region(PFN_DOWN(paddr
),
278 paddr
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
279 if (ei
->type
!= E820_RAM
)
280 register_nosave_region(PFN_UP(ei
->addr
),
283 if (paddr
>= (end_pfn
<< PAGE_SHIFT
))
289 * Finds an active region in the address range from start_pfn to end_pfn and
290 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
292 static int __init
e820_find_active_region(const struct e820entry
*ei
,
293 unsigned long start_pfn
,
294 unsigned long end_pfn
,
295 unsigned long *ei_startpfn
,
296 unsigned long *ei_endpfn
)
298 *ei_startpfn
= round_up(ei
->addr
, PAGE_SIZE
) >> PAGE_SHIFT
;
299 *ei_endpfn
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
) >> PAGE_SHIFT
;
301 /* Skip map entries smaller than a page */
302 if (*ei_startpfn
>= *ei_endpfn
)
305 /* Check if end_pfn_map should be updated */
306 if (ei
->type
!= E820_RAM
&& *ei_endpfn
> end_pfn_map
)
307 end_pfn_map
= *ei_endpfn
;
309 /* Skip if map is outside the node */
310 if (ei
->type
!= E820_RAM
|| *ei_endpfn
<= start_pfn
||
311 *ei_startpfn
>= end_pfn
)
314 /* Check for overlaps */
315 if (*ei_startpfn
< start_pfn
)
316 *ei_startpfn
= start_pfn
;
317 if (*ei_endpfn
> end_pfn
)
318 *ei_endpfn
= end_pfn
;
320 /* Obey end_user_pfn to save on memmap */
321 if (*ei_startpfn
>= end_user_pfn
)
323 if (*ei_endpfn
> end_user_pfn
)
324 *ei_endpfn
= end_user_pfn
;
329 /* Walk the e820 map and register active regions within a node */
331 e820_register_active_regions(int nid
, unsigned long start_pfn
,
332 unsigned long end_pfn
)
334 unsigned long ei_startpfn
;
335 unsigned long ei_endpfn
;
338 for (i
= 0; i
< e820
.nr_map
; i
++)
339 if (e820_find_active_region(&e820
.map
[i
],
341 &ei_startpfn
, &ei_endpfn
))
342 add_active_range(nid
, ei_startpfn
, ei_endpfn
);
346 * Add a memory region to the kernel e820 map.
348 void __init
add_memory_region(unsigned long start
, unsigned long size
, int type
)
353 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
357 e820
.map
[x
].addr
= start
;
358 e820
.map
[x
].size
= size
;
359 e820
.map
[x
].type
= type
;
364 * Find the hole size (in bytes) in the memory range.
365 * @start: starting address of the memory range to scan
366 * @end: ending address of the memory range to scan
368 unsigned long __init
e820_hole_size(unsigned long start
, unsigned long end
)
370 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
371 unsigned long end_pfn
= end
>> PAGE_SHIFT
;
372 unsigned long ei_startpfn
, ei_endpfn
, ram
= 0;
375 for (i
= 0; i
< e820
.nr_map
; i
++) {
376 if (e820_find_active_region(&e820
.map
[i
],
378 &ei_startpfn
, &ei_endpfn
))
379 ram
+= ei_endpfn
- ei_startpfn
;
381 return end
- start
- (ram
<< PAGE_SHIFT
);
384 static void __init
e820_print_map(char *who
)
388 for (i
= 0; i
< e820
.nr_map
; i
++) {
389 printk(KERN_INFO
" %s: %016Lx - %016Lx ", who
,
390 (unsigned long long) e820
.map
[i
].addr
,
392 (e820
.map
[i
].addr
+ e820
.map
[i
].size
));
393 switch (e820
.map
[i
].type
) {
395 printk(KERN_CONT
"(usable)\n");
398 printk(KERN_CONT
"(reserved)\n");
401 printk(KERN_CONT
"(ACPI data)\n");
404 printk(KERN_CONT
"(ACPI NVS)\n");
407 printk(KERN_CONT
"type %u\n", e820
.map
[i
].type
);
414 * Sanitize the BIOS e820 map.
416 * Some e820 responses include overlapping entries. The following
417 * replaces the original e820 map with a new one, removing overlaps.
420 static int __init
sanitize_e820_map(struct e820entry
*biosmap
, char *pnr_map
)
422 struct change_member
{
423 struct e820entry
*pbios
; /* pointer to original bios entry */
424 unsigned long long addr
; /* address for this change point */
426 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
427 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
428 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
429 static struct e820entry new_bios
[E820MAX
] __initdata
;
430 struct change_member
*change_tmp
;
431 unsigned long current_type
, last_type
;
432 unsigned long long last_addr
;
433 int chgidx
, still_changing
;
436 int old_nr
, new_nr
, chg_nr
;
440 Visually we're performing the following
441 (1,2,3,4 = memory types)...
443 Sample memory map (w/overlaps):
444 ____22__________________
445 ______________________4_
446 ____1111________________
447 _44_____________________
448 11111111________________
449 ____________________33__
450 ___________44___________
451 __________33333_________
452 ______________22________
453 ___________________2222_
454 _________111111111______
455 _____________________11_
456 _________________4______
458 Sanitized equivalent (no overlap):
459 1_______________________
460 _44_____________________
461 ___1____________________
462 ____22__________________
463 ______11________________
464 _________1______________
465 __________3_____________
466 ___________44___________
467 _____________33_________
468 _______________2________
469 ________________1_______
470 _________________4______
471 ___________________2____
472 ____________________33__
473 ______________________4_
476 /* if there's only one memory region, don't bother */
482 /* bail out if we find any unreasonable addresses in bios map */
483 for (i
= 0; i
< old_nr
; i
++)
484 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
)
487 /* create pointers for initial change-point information (for sorting) */
488 for (i
= 0; i
< 2 * old_nr
; i
++)
489 change_point
[i
] = &change_point_list
[i
];
491 /* record all known change-points (starting and ending addresses),
492 omitting those that are for empty memory regions */
494 for (i
= 0; i
< old_nr
; i
++) {
495 if (biosmap
[i
].size
!= 0) {
496 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
497 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
498 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+
500 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
505 /* sort change-point list by memory addresses (low -> high) */
507 while (still_changing
) {
509 for (i
= 1; i
< chg_nr
; i
++) {
510 unsigned long long curaddr
, lastaddr
;
511 unsigned long long curpbaddr
, lastpbaddr
;
513 curaddr
= change_point
[i
]->addr
;
514 lastaddr
= change_point
[i
- 1]->addr
;
515 curpbaddr
= change_point
[i
]->pbios
->addr
;
516 lastpbaddr
= change_point
[i
- 1]->pbios
->addr
;
519 * swap entries, when:
521 * curaddr > lastaddr or
522 * curaddr == lastaddr and curaddr == curpbaddr and
523 * lastaddr != lastpbaddr
525 if (curaddr
< lastaddr
||
526 (curaddr
== lastaddr
&& curaddr
== curpbaddr
&&
527 lastaddr
!= lastpbaddr
)) {
528 change_tmp
= change_point
[i
];
529 change_point
[i
] = change_point
[i
-1];
530 change_point
[i
-1] = change_tmp
;
536 /* create a new bios memory map, removing overlaps */
537 overlap_entries
= 0; /* number of entries in the overlap table */
538 new_bios_entry
= 0; /* index for creating new bios map entries */
539 last_type
= 0; /* start with undefined memory type */
540 last_addr
= 0; /* start with 0 as last starting address */
542 /* loop through change-points, determining affect on the new bios map */
543 for (chgidx
= 0; chgidx
< chg_nr
; chgidx
++) {
544 /* keep track of all overlapping bios entries */
545 if (change_point
[chgidx
]->addr
==
546 change_point
[chgidx
]->pbios
->addr
) {
548 * add map entry to overlap list (> 1 entry
549 * implies an overlap)
551 overlap_list
[overlap_entries
++] =
552 change_point
[chgidx
]->pbios
;
555 * remove entry from list (order independent,
558 for (i
= 0; i
< overlap_entries
; i
++) {
559 if (overlap_list
[i
] ==
560 change_point
[chgidx
]->pbios
)
562 overlap_list
[overlap_entries
-1];
567 * if there are overlapping entries, decide which
568 * "type" to use (larger value takes precedence --
569 * 1=usable, 2,3,4,4+=unusable)
572 for (i
= 0; i
< overlap_entries
; i
++)
573 if (overlap_list
[i
]->type
> current_type
)
574 current_type
= overlap_list
[i
]->type
;
576 * continue building up new bios map based on this
579 if (current_type
!= last_type
) {
580 if (last_type
!= 0) {
581 new_bios
[new_bios_entry
].size
=
582 change_point
[chgidx
]->addr
- last_addr
;
584 * move forward only if the new size
587 if (new_bios
[new_bios_entry
].size
!= 0)
589 * no more space left for new
592 if (++new_bios_entry
>= E820MAX
)
595 if (current_type
!= 0) {
596 new_bios
[new_bios_entry
].addr
=
597 change_point
[chgidx
]->addr
;
598 new_bios
[new_bios_entry
].type
= current_type
;
599 last_addr
= change_point
[chgidx
]->addr
;
601 last_type
= current_type
;
604 /* retain count for new bios entries */
605 new_nr
= new_bios_entry
;
607 /* copy new bios mapping into original location */
608 memcpy(biosmap
, new_bios
, new_nr
* sizeof(struct e820entry
));
615 * Copy the BIOS e820 map into a safe place.
617 * Sanity-check it while we're at it..
619 * If we're lucky and live on a modern system, the setup code
620 * will have given us a memory map that we can use to properly
621 * set up memory. If we aren't, we'll fake a memory map.
623 static int __init
copy_e820_map(struct e820entry
*biosmap
, int nr_map
)
625 /* Only one memory region (or negative)? Ignore it */
630 unsigned long start
= biosmap
->addr
;
631 unsigned long size
= biosmap
->size
;
632 unsigned long end
= start
+ size
;
633 unsigned long type
= biosmap
->type
;
635 /* Overflow in 64 bits? Ignore the memory map. */
639 add_memory_region(start
, size
, type
);
640 } while (biosmap
++, --nr_map
);
644 static void early_panic(char *msg
)
650 /* We're not void only for x86 32-bit compat */
651 char * __init
machine_specific_memory_setup(void)
653 char *who
= "BIOS-e820";
655 * Try to copy the BIOS-supplied E820-map.
657 * Otherwise fake a memory map; one section from 0k->640k,
658 * the next section from 1mb->appropriate_mem_k
660 sanitize_e820_map(boot_params
.e820_map
, &boot_params
.e820_entries
);
661 if (copy_e820_map(boot_params
.e820_map
, boot_params
.e820_entries
) < 0)
662 early_panic("Cannot find a valid memory map");
663 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
666 /* In case someone cares... */
670 static int __init
parse_memopt(char *p
)
674 end_user_pfn
= memparse(p
, &p
);
675 end_user_pfn
>>= PAGE_SHIFT
;
678 early_param("mem", parse_memopt
);
680 static int userdef __initdata
;
682 static int __init
parse_memmap_opt(char *p
)
685 unsigned long long start_at
, mem_size
;
687 if (!strcmp(p
, "exactmap")) {
688 #ifdef CONFIG_CRASH_DUMP
690 * If we are doing a crash dump, we still need to know
691 * the real mem size before original memory map is
694 e820_register_active_regions(0, 0, -1UL);
695 saved_max_pfn
= e820_end_of_ram();
696 remove_all_active_ranges();
705 mem_size
= memparse(p
, &p
);
711 start_at
= memparse(p
+1, &p
);
712 add_memory_region(start_at
, mem_size
, E820_RAM
);
713 } else if (*p
== '#') {
714 start_at
= memparse(p
+1, &p
);
715 add_memory_region(start_at
, mem_size
, E820_ACPI
);
716 } else if (*p
== '$') {
717 start_at
= memparse(p
+1, &p
);
718 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
720 end_user_pfn
= (mem_size
>> PAGE_SHIFT
);
722 return *p
== '\0' ? 0 : -EINVAL
;
724 early_param("memmap", parse_memmap_opt
);
726 void __init
finish_e820_parsing(void)
729 char nr
= e820
.nr_map
;
731 if (sanitize_e820_map(e820
.map
, &nr
) < 0)
732 early_panic("Invalid user supplied memory map");
735 printk(KERN_INFO
"user-defined physical RAM map:\n");
736 e820_print_map("user");
740 void __init
update_e820(void)
744 nr_map
= e820
.nr_map
;
745 if (sanitize_e820_map(e820
.map
, &nr_map
))
747 e820
.nr_map
= nr_map
;
748 printk(KERN_INFO
"modified physical RAM map:\n");
749 e820_print_map("modified");
752 unsigned long pci_mem_start
= 0xaeedbabe;
753 EXPORT_SYMBOL(pci_mem_start
);
756 * Search for the biggest gap in the low 32 bits of the e820
757 * memory space. We pass this space to PCI to assign MMIO resources
758 * for hotplug or unconfigured devices in.
759 * Hopefully the BIOS let enough space left.
761 __init
void e820_setup_gap(void)
763 unsigned long gapstart
, gapsize
, round
;
768 last
= 0x100000000ull
;
769 gapstart
= 0x10000000;
773 unsigned long long start
= e820
.map
[i
].addr
;
774 unsigned long long end
= start
+ e820
.map
[i
].size
;
777 * Since "last" is at most 4GB, we know we'll
778 * fit in 32 bits if this condition is true
781 unsigned long gap
= last
- end
;
794 gapstart
= (end_pfn
<< PAGE_SHIFT
) + 1024*1024;
795 printk(KERN_ERR
"PCI: Warning: Cannot find a gap in the 32bit "
797 KERN_ERR
"PCI: Unassigned devices with 32bit resource "
798 "registers may break!\n");
802 * See how much we want to round up: start off with
803 * rounding to the next 1MB area.
806 while ((gapsize
>> 4) > round
)
808 /* Fun with two's complement */
809 pci_mem_start
= (gapstart
+ round
) & -round
;
812 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
813 pci_mem_start
, gapstart
, gapsize
);
816 int __init
arch_get_ram_range(int slot
, u64
*addr
, u64
*size
)
820 if (slot
< 0 || slot
>= e820
.nr_map
)
822 for (i
= slot
; i
< e820
.nr_map
; i
++) {
823 if (e820
.map
[i
].type
!= E820_RAM
)
827 if (i
== e820
.nr_map
|| e820
.map
[i
].addr
> (max_pfn
<< PAGE_SHIFT
))
829 *addr
= e820
.map
[i
].addr
;
830 *size
= min_t(u64
, e820
.map
[i
].size
+ e820
.map
[i
].addr
,
831 max_pfn
<< PAGE_SHIFT
) - *addr
;