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>
33 * PFN of last memory page.
35 unsigned long end_pfn
;
36 EXPORT_SYMBOL(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
;
50 extern struct resource code_resource
, data_resource
;
52 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
53 static inline int bad_addr(unsigned long *addrp
, unsigned long size
)
55 unsigned long addr
= *addrp
, last
= addr
+ size
;
57 /* various gunk below that needed for SMP startup */
59 *addrp
= PAGE_ALIGN(0x8000);
63 /* direct mapping tables of the kernel */
64 if (last
>= table_start
<<PAGE_SHIFT
&& addr
< table_end
<<PAGE_SHIFT
) {
65 *addrp
= PAGE_ALIGN(table_end
<< PAGE_SHIFT
);
70 #ifdef CONFIG_BLK_DEV_INITRD
71 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
) {
72 unsigned long ramdisk_image
= boot_params
.hdr
.ramdisk_image
;
73 unsigned long ramdisk_size
= boot_params
.hdr
.ramdisk_size
;
74 unsigned long ramdisk_end
= ramdisk_image
+ramdisk_size
;
76 if (last
>= ramdisk_image
&& addr
< ramdisk_end
) {
77 *addrp
= PAGE_ALIGN(ramdisk_end
);
83 if (last
>= __pa_symbol(&_text
) && addr
< __pa_symbol(&_end
)) {
84 *addrp
= PAGE_ALIGN(__pa_symbol(&_end
));
88 if (last
>= ebda_addr
&& addr
< ebda_addr
+ ebda_size
) {
89 *addrp
= PAGE_ALIGN(ebda_addr
+ ebda_size
);
94 /* NUMA memory to node map */
95 if (last
>= nodemap_addr
&& addr
< nodemap_addr
+ nodemap_size
) {
96 *addrp
= nodemap_addr
+ nodemap_size
;
100 /* XXX ramdisk image here? */
105 * This function checks if any part of the range <start,end> is mapped
109 e820_any_mapped(unsigned long start
, unsigned long end
, unsigned type
)
112 for (i
= 0; i
< e820
.nr_map
; i
++) {
113 struct e820entry
*ei
= &e820
.map
[i
];
114 if (type
&& ei
->type
!= type
)
116 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
122 EXPORT_SYMBOL_GPL(e820_any_mapped
);
125 * This function checks if the entire range <start,end> is mapped with type.
127 * Note: this function only works correct if the e820 table is sorted and
128 * not-overlapping, which is the case
130 int __init
e820_all_mapped(unsigned long start
, unsigned long end
, unsigned type
)
133 for (i
= 0; i
< e820
.nr_map
; i
++) {
134 struct e820entry
*ei
= &e820
.map
[i
];
135 if (type
&& ei
->type
!= type
)
137 /* is the region (part) in overlap with the current region ?*/
138 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
141 /* if the region is at the beginning of <start,end> we move
142 * start to the end of the region since it's ok until there
144 if (ei
->addr
<= start
)
145 start
= ei
->addr
+ ei
->size
;
146 /* if start is now at or beyond end, we're done, full coverage */
148 return 1; /* we're done */
154 * Find a free area in a specific range.
156 unsigned long __init
find_e820_area(unsigned long start
, unsigned long end
, unsigned size
)
159 for (i
= 0; i
< e820
.nr_map
; i
++) {
160 struct e820entry
*ei
= &e820
.map
[i
];
161 unsigned long addr
= ei
->addr
, last
;
162 if (ei
->type
!= E820_RAM
)
166 if (addr
> ei
->addr
+ ei
->size
)
168 while (bad_addr(&addr
, size
) && addr
+size
<= ei
->addr
+ei
->size
)
170 last
= PAGE_ALIGN(addr
) + size
;
171 if (last
> ei
->addr
+ ei
->size
)
181 * Find the highest page frame number we have available
183 unsigned long __init
e820_end_of_ram(void)
185 unsigned long end_pfn
= 0;
186 end_pfn
= find_max_pfn_with_active_regions();
188 if (end_pfn
> end_pfn_map
)
189 end_pfn_map
= end_pfn
;
190 if (end_pfn_map
> MAXMEM
>>PAGE_SHIFT
)
191 end_pfn_map
= MAXMEM
>>PAGE_SHIFT
;
192 if (end_pfn
> end_user_pfn
)
193 end_pfn
= end_user_pfn
;
194 if (end_pfn
> end_pfn_map
)
195 end_pfn
= end_pfn_map
;
197 printk("end_pfn_map = %lu\n", end_pfn_map
);
202 * Mark e820 reserved areas as busy for the resource manager.
204 void __init
e820_reserve_resources(void)
207 for (i
= 0; i
< e820
.nr_map
; i
++) {
208 struct resource
*res
;
209 res
= alloc_bootmem_low(sizeof(struct resource
));
210 switch (e820
.map
[i
].type
) {
211 case E820_RAM
: res
->name
= "System RAM"; break;
212 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
213 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
214 default: res
->name
= "reserved";
216 res
->start
= e820
.map
[i
].addr
;
217 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
218 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
219 request_resource(&iomem_resource
, res
);
220 if (e820
.map
[i
].type
== E820_RAM
) {
222 * We don't know which RAM region contains kernel data,
223 * so we try it repeatedly and let the resource manager
226 request_resource(res
, &code_resource
);
227 request_resource(res
, &data_resource
);
229 if (crashk_res
.start
!= crashk_res
.end
)
230 request_resource(res
, &crashk_res
);
237 * Find the ranges of physical addresses that do not correspond to
238 * e820 RAM areas and mark the corresponding pages as nosave for software
239 * suspend and suspend to RAM.
241 * This function requires the e820 map to be sorted and without any
242 * overlapping entries and assumes the first e820 area to be RAM.
244 void __init
e820_mark_nosave_regions(void)
249 paddr
= round_down(e820
.map
[0].addr
+ e820
.map
[0].size
, PAGE_SIZE
);
250 for (i
= 1; i
< e820
.nr_map
; i
++) {
251 struct e820entry
*ei
= &e820
.map
[i
];
253 if (paddr
< ei
->addr
)
254 register_nosave_region(PFN_DOWN(paddr
),
257 paddr
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
258 if (ei
->type
!= E820_RAM
)
259 register_nosave_region(PFN_UP(ei
->addr
),
262 if (paddr
>= (end_pfn
<< PAGE_SHIFT
))
268 * Finds an active region in the address range from start_pfn to end_pfn and
269 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
271 static int __init
e820_find_active_region(const struct e820entry
*ei
,
272 unsigned long start_pfn
,
273 unsigned long end_pfn
,
274 unsigned long *ei_startpfn
,
275 unsigned long *ei_endpfn
)
277 *ei_startpfn
= round_up(ei
->addr
, PAGE_SIZE
) >> PAGE_SHIFT
;
278 *ei_endpfn
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
) >> PAGE_SHIFT
;
280 /* Skip map entries smaller than a page */
281 if (*ei_startpfn
>= *ei_endpfn
)
284 /* Check if end_pfn_map should be updated */
285 if (ei
->type
!= E820_RAM
&& *ei_endpfn
> end_pfn_map
)
286 end_pfn_map
= *ei_endpfn
;
288 /* Skip if map is outside the node */
289 if (ei
->type
!= E820_RAM
|| *ei_endpfn
<= start_pfn
||
290 *ei_startpfn
>= end_pfn
)
293 /* Check for overlaps */
294 if (*ei_startpfn
< start_pfn
)
295 *ei_startpfn
= start_pfn
;
296 if (*ei_endpfn
> end_pfn
)
297 *ei_endpfn
= end_pfn
;
299 /* Obey end_user_pfn to save on memmap */
300 if (*ei_startpfn
>= end_user_pfn
)
302 if (*ei_endpfn
> end_user_pfn
)
303 *ei_endpfn
= end_user_pfn
;
308 /* Walk the e820 map and register active regions within a node */
310 e820_register_active_regions(int nid
, unsigned long start_pfn
,
311 unsigned long end_pfn
)
313 unsigned long ei_startpfn
;
314 unsigned long ei_endpfn
;
317 for (i
= 0; i
< e820
.nr_map
; i
++)
318 if (e820_find_active_region(&e820
.map
[i
],
320 &ei_startpfn
, &ei_endpfn
))
321 add_active_range(nid
, ei_startpfn
, ei_endpfn
);
325 * Add a memory region to the kernel e820 map.
327 void __init
add_memory_region(unsigned long start
, unsigned long size
, int type
)
332 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
336 e820
.map
[x
].addr
= start
;
337 e820
.map
[x
].size
= size
;
338 e820
.map
[x
].type
= type
;
343 * Find the hole size (in bytes) in the memory range.
344 * @start: starting address of the memory range to scan
345 * @end: ending address of the memory range to scan
347 unsigned long __init
e820_hole_size(unsigned long start
, unsigned long end
)
349 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
350 unsigned long end_pfn
= end
>> PAGE_SHIFT
;
351 unsigned long ei_startpfn
;
352 unsigned long ei_endpfn
;
353 unsigned long ram
= 0;
356 for (i
= 0; i
< e820
.nr_map
; i
++) {
357 if (e820_find_active_region(&e820
.map
[i
],
359 &ei_startpfn
, &ei_endpfn
))
360 ram
+= ei_endpfn
- ei_startpfn
;
362 return end
- start
- (ram
<< PAGE_SHIFT
);
365 void __init
e820_print_map(char *who
)
369 for (i
= 0; i
< e820
.nr_map
; i
++) {
370 printk(KERN_INFO
" %s: %016Lx - %016Lx ", who
,
371 (unsigned long long) e820
.map
[i
].addr
,
372 (unsigned long long) (e820
.map
[i
].addr
+ e820
.map
[i
].size
));
373 switch (e820
.map
[i
].type
) {
374 case E820_RAM
: printk("(usable)\n");
377 printk("(reserved)\n");
380 printk("(ACPI data)\n");
383 printk("(ACPI NVS)\n");
385 default: printk("type %u\n", e820
.map
[i
].type
);
392 * Sanitize the BIOS e820 map.
394 * Some e820 responses include overlapping entries. The following
395 * replaces the original e820 map with a new one, removing overlaps.
398 static int __init
sanitize_e820_map(struct e820entry
* biosmap
, char * pnr_map
)
400 struct change_member
{
401 struct e820entry
*pbios
; /* pointer to original bios entry */
402 unsigned long long addr
; /* address for this change point */
404 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
405 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
406 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
407 static struct e820entry new_bios
[E820MAX
] __initdata
;
408 struct change_member
*change_tmp
;
409 unsigned long current_type
, last_type
;
410 unsigned long long last_addr
;
411 int chgidx
, still_changing
;
414 int old_nr
, new_nr
, chg_nr
;
418 Visually we're performing the following (1,2,3,4 = memory types)...
420 Sample memory map (w/overlaps):
421 ____22__________________
422 ______________________4_
423 ____1111________________
424 _44_____________________
425 11111111________________
426 ____________________33__
427 ___________44___________
428 __________33333_________
429 ______________22________
430 ___________________2222_
431 _________111111111______
432 _____________________11_
433 _________________4______
435 Sanitized equivalent (no overlap):
436 1_______________________
437 _44_____________________
438 ___1____________________
439 ____22__________________
440 ______11________________
441 _________1______________
442 __________3_____________
443 ___________44___________
444 _____________33_________
445 _______________2________
446 ________________1_______
447 _________________4______
448 ___________________2____
449 ____________________33__
450 ______________________4_
453 /* if there's only one memory region, don't bother */
459 /* bail out if we find any unreasonable addresses in bios map */
460 for (i
=0; i
<old_nr
; i
++)
461 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
)
464 /* create pointers for initial change-point information (for sorting) */
465 for (i
=0; i
< 2*old_nr
; i
++)
466 change_point
[i
] = &change_point_list
[i
];
468 /* record all known change-points (starting and ending addresses),
469 omitting those that are for empty memory regions */
471 for (i
=0; i
< old_nr
; i
++) {
472 if (biosmap
[i
].size
!= 0) {
473 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
474 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
475 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+ biosmap
[i
].size
;
476 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
481 /* sort change-point list by memory addresses (low -> high) */
483 while (still_changing
) {
485 for (i
=1; i
< chg_nr
; i
++) {
486 /* if <current_addr> > <last_addr>, swap */
487 /* or, if current=<start_addr> & last=<end_addr>, swap */
488 if ((change_point
[i
]->addr
< change_point
[i
-1]->addr
) ||
489 ((change_point
[i
]->addr
== change_point
[i
-1]->addr
) &&
490 (change_point
[i
]->addr
== change_point
[i
]->pbios
->addr
) &&
491 (change_point
[i
-1]->addr
!= change_point
[i
-1]->pbios
->addr
))
494 change_tmp
= change_point
[i
];
495 change_point
[i
] = change_point
[i
-1];
496 change_point
[i
-1] = change_tmp
;
502 /* create a new bios memory map, removing overlaps */
503 overlap_entries
=0; /* number of entries in the overlap table */
504 new_bios_entry
=0; /* index for creating new bios map entries */
505 last_type
= 0; /* start with undefined memory type */
506 last_addr
= 0; /* start with 0 as last starting address */
507 /* loop through change-points, determining affect on the new bios map */
508 for (chgidx
=0; chgidx
< chg_nr
; chgidx
++)
510 /* keep track of all overlapping bios entries */
511 if (change_point
[chgidx
]->addr
== change_point
[chgidx
]->pbios
->addr
)
513 /* add map entry to overlap list (> 1 entry implies an overlap) */
514 overlap_list
[overlap_entries
++]=change_point
[chgidx
]->pbios
;
518 /* remove entry from list (order independent, so swap with last) */
519 for (i
=0; i
<overlap_entries
; i
++)
521 if (overlap_list
[i
] == change_point
[chgidx
]->pbios
)
522 overlap_list
[i
] = overlap_list
[overlap_entries
-1];
526 /* if there are overlapping entries, decide which "type" to use */
527 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
529 for (i
=0; i
<overlap_entries
; i
++)
530 if (overlap_list
[i
]->type
> current_type
)
531 current_type
= overlap_list
[i
]->type
;
532 /* continue building up new bios map based on this information */
533 if (current_type
!= last_type
) {
534 if (last_type
!= 0) {
535 new_bios
[new_bios_entry
].size
=
536 change_point
[chgidx
]->addr
- last_addr
;
537 /* move forward only if the new size was non-zero */
538 if (new_bios
[new_bios_entry
].size
!= 0)
539 if (++new_bios_entry
>= E820MAX
)
540 break; /* no more space left for new bios entries */
542 if (current_type
!= 0) {
543 new_bios
[new_bios_entry
].addr
= change_point
[chgidx
]->addr
;
544 new_bios
[new_bios_entry
].type
= current_type
;
545 last_addr
=change_point
[chgidx
]->addr
;
547 last_type
= current_type
;
550 new_nr
= new_bios_entry
; /* retain count for new bios entries */
552 /* copy new bios mapping into original location */
553 memcpy(biosmap
, new_bios
, new_nr
*sizeof(struct e820entry
));
560 * Copy the BIOS e820 map into a safe place.
562 * Sanity-check it while we're at it..
564 * If we're lucky and live on a modern system, the setup code
565 * will have given us a memory map that we can use to properly
566 * set up memory. If we aren't, we'll fake a memory map.
568 static int __init
copy_e820_map(struct e820entry
* biosmap
, int nr_map
)
570 /* Only one memory region (or negative)? Ignore it */
575 unsigned long start
= biosmap
->addr
;
576 unsigned long size
= biosmap
->size
;
577 unsigned long end
= start
+ size
;
578 unsigned long type
= biosmap
->type
;
580 /* Overflow in 64 bits? Ignore the memory map. */
584 add_memory_region(start
, size
, type
);
585 } while (biosmap
++,--nr_map
);
589 void early_panic(char *msg
)
595 void __init
setup_memory_region(void)
598 * Try to copy the BIOS-supplied E820-map.
600 * Otherwise fake a memory map; one section from 0k->640k,
601 * the next section from 1mb->appropriate_mem_k
603 sanitize_e820_map(boot_params
.e820_map
, &boot_params
.e820_entries
);
604 if (copy_e820_map(boot_params
.e820_map
, boot_params
.e820_entries
) < 0)
605 early_panic("Cannot find a valid memory map");
606 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
607 e820_print_map("BIOS-e820");
610 static int __init
parse_memopt(char *p
)
614 end_user_pfn
= memparse(p
, &p
);
615 end_user_pfn
>>= PAGE_SHIFT
;
618 early_param("mem", parse_memopt
);
620 static int userdef __initdata
;
622 static int __init
parse_memmap_opt(char *p
)
625 unsigned long long start_at
, mem_size
;
627 if (!strcmp(p
, "exactmap")) {
628 #ifdef CONFIG_CRASH_DUMP
629 /* If we are doing a crash dump, we
630 * still need to know the real mem
631 * size before original memory map is
634 e820_register_active_regions(0, 0, -1UL);
635 saved_max_pfn
= e820_end_of_ram();
636 remove_all_active_ranges();
645 mem_size
= memparse(p
, &p
);
649 start_at
= memparse(p
+1, &p
);
650 add_memory_region(start_at
, mem_size
, E820_RAM
);
651 } else if (*p
== '#') {
652 start_at
= memparse(p
+1, &p
);
653 add_memory_region(start_at
, mem_size
, E820_ACPI
);
654 } else if (*p
== '$') {
655 start_at
= memparse(p
+1, &p
);
656 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
658 end_user_pfn
= (mem_size
>> PAGE_SHIFT
);
660 return *p
== '\0' ? 0 : -EINVAL
;
662 early_param("memmap", parse_memmap_opt
);
664 void __init
finish_e820_parsing(void)
667 printk(KERN_INFO
"user-defined physical RAM map:\n");
668 e820_print_map("user");
672 unsigned long pci_mem_start
= 0xaeedbabe;
673 EXPORT_SYMBOL(pci_mem_start
);
676 * Search for the biggest gap in the low 32 bits of the e820
677 * memory space. We pass this space to PCI to assign MMIO resources
678 * for hotplug or unconfigured devices in.
679 * Hopefully the BIOS let enough space left.
681 __init
void e820_setup_gap(void)
683 unsigned long gapstart
, gapsize
, round
;
688 last
= 0x100000000ull
;
689 gapstart
= 0x10000000;
693 unsigned long long start
= e820
.map
[i
].addr
;
694 unsigned long long end
= start
+ e820
.map
[i
].size
;
697 * Since "last" is at most 4GB, we know we'll
698 * fit in 32 bits if this condition is true
701 unsigned long gap
= last
- end
;
714 gapstart
= (end_pfn
<< PAGE_SHIFT
) + 1024*1024;
715 printk(KERN_ERR
"PCI: Warning: Cannot find a gap in the 32bit address range\n"
716 KERN_ERR
"PCI: Unassigned devices with 32bit resource registers may break!\n");
720 * See how much we want to round up: start off with
721 * rounding to the next 1MB area.
724 while ((gapsize
>> 4) > round
)
726 /* Fun with two's complement */
727 pci_mem_start
= (gapstart
+ round
) & -round
;
729 printk(KERN_INFO
"Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
730 pci_mem_start
, gapstart
, gapsize
);
733 int __init
arch_get_ram_range(int slot
, u64
*addr
, u64
*size
)
737 if (slot
< 0 || slot
>= e820
.nr_map
)
739 for (i
= slot
; i
< e820
.nr_map
; i
++) {
740 if (e820
.map
[i
].type
!= E820_RAM
)
744 if (i
== e820
.nr_map
|| e820
.map
[i
].addr
> (max_pfn
<< PAGE_SHIFT
))
746 *addr
= e820
.map
[i
].addr
;
747 *size
= min_t(u64
, e820
.map
[i
].size
+ e820
.map
[i
].addr
,
748 max_pfn
<< PAGE_SHIFT
) - *addr
;