2 * Handle the memory map.
3 * The functions here do the job until bootmem takes over.
4 * $Id: e820.c,v 1.4 2002/09/19 19:25:32 ak Exp $
6 * Getting sanitize_e820_map() in sync with i386 version by applying change:
7 * - Provisions for empty E820 memory regions (reported by certain BIOSes).
8 * Alex Achenbach <xela@slit.de>, December 2002.
9 * Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
12 #include <linux/config.h>
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/init.h>
16 #include <linux/bootmem.h>
17 #include <linux/ioport.h>
18 #include <linux/string.h>
19 #include <linux/kexec.h>
20 #include <linux/module.h>
24 #include <asm/proto.h>
25 #include <asm/bootsetup.h>
26 #include <asm/sections.h>
29 * PFN of last memory page.
31 unsigned long end_pfn
;
32 EXPORT_SYMBOL(end_pfn
);
35 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
36 * The direct mapping extends to end_pfn_map, so that we can directly access
37 * apertures, ACPI and other tables without having to play with fixmaps.
39 unsigned long end_pfn_map
;
42 * Last pfn which the user wants to use.
44 unsigned long end_user_pfn
= MAXMEM
>>PAGE_SHIFT
;
46 extern struct resource code_resource
, data_resource
;
48 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
49 static inline int bad_addr(unsigned long *addrp
, unsigned long size
)
51 unsigned long addr
= *addrp
, last
= addr
+ size
;
53 /* various gunk below that needed for SMP startup */
59 /* direct mapping tables of the kernel */
60 if (last
>= table_start
<<PAGE_SHIFT
&& addr
< table_end
<<PAGE_SHIFT
) {
61 *addrp
= table_end
<< PAGE_SHIFT
;
66 #ifdef CONFIG_BLK_DEV_INITRD
67 if (LOADER_TYPE
&& INITRD_START
&& last
>= INITRD_START
&&
68 addr
< INITRD_START
+INITRD_SIZE
) {
69 *addrp
= INITRD_START
+ INITRD_SIZE
;
73 /* kernel code + 640k memory hole (later should not be needed, but
74 be paranoid for now) */
75 if (last
>= 640*1024 && addr
< __pa_symbol(&_end
)) {
76 *addrp
= __pa_symbol(&_end
);
79 /* XXX ramdisk image here? */
84 e820_any_mapped(unsigned long start
, unsigned long end
, unsigned type
)
87 for (i
= 0; i
< e820
.nr_map
; i
++) {
88 struct e820entry
*ei
= &e820
.map
[i
];
89 if (type
&& ei
->type
!= type
)
91 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
99 * Find a free area in a specific range.
101 unsigned long __init
find_e820_area(unsigned long start
, unsigned long end
, unsigned size
)
104 for (i
= 0; i
< e820
.nr_map
; i
++) {
105 struct e820entry
*ei
= &e820
.map
[i
];
106 unsigned long addr
= ei
->addr
, last
;
107 if (ei
->type
!= E820_RAM
)
111 if (addr
> ei
->addr
+ ei
->size
)
113 while (bad_addr(&addr
, size
) && addr
+size
< ei
->addr
+ ei
->size
)
116 if (last
> ei
->addr
+ ei
->size
)
126 * Free bootmem based on the e820 table for a node.
128 void __init
e820_bootmem_free(pg_data_t
*pgdat
, unsigned long start
,unsigned long end
)
131 for (i
= 0; i
< e820
.nr_map
; i
++) {
132 struct e820entry
*ei
= &e820
.map
[i
];
133 unsigned long last
, addr
;
135 if (ei
->type
!= E820_RAM
||
136 ei
->addr
+ei
->size
<= start
||
140 addr
= round_up(ei
->addr
, PAGE_SIZE
);
144 last
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
148 if (last
> addr
&& last
-addr
>= PAGE_SIZE
)
149 free_bootmem_node(pgdat
, addr
, last
-addr
);
154 * Find the highest page frame number we have available
156 unsigned long __init
e820_end_of_ram(void)
159 unsigned long end_pfn
= 0;
161 for (i
= 0; i
< e820
.nr_map
; i
++) {
162 struct e820entry
*ei
= &e820
.map
[i
];
163 unsigned long start
, end
;
165 start
= round_up(ei
->addr
, PAGE_SIZE
);
166 end
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
169 if (ei
->type
== E820_RAM
) {
170 if (end
> end_pfn
<<PAGE_SHIFT
)
171 end_pfn
= end
>>PAGE_SHIFT
;
173 if (end
> end_pfn_map
<<PAGE_SHIFT
)
174 end_pfn_map
= end
>>PAGE_SHIFT
;
178 if (end_pfn
> end_pfn_map
)
179 end_pfn_map
= end_pfn
;
180 if (end_pfn_map
> MAXMEM
>>PAGE_SHIFT
)
181 end_pfn_map
= MAXMEM
>>PAGE_SHIFT
;
182 if (end_pfn
> end_user_pfn
)
183 end_pfn
= end_user_pfn
;
184 if (end_pfn
> end_pfn_map
)
185 end_pfn
= end_pfn_map
;
191 * Compute how much memory is missing in a range.
192 * Unlike the other functions in this file the arguments are in page numbers.
195 e820_hole_size(unsigned long start_pfn
, unsigned long end_pfn
)
197 unsigned long ram
= 0;
198 unsigned long start
= start_pfn
<< PAGE_SHIFT
;
199 unsigned long end
= end_pfn
<< PAGE_SHIFT
;
201 for (i
= 0; i
< e820
.nr_map
; i
++) {
202 struct e820entry
*ei
= &e820
.map
[i
];
203 unsigned long last
, addr
;
205 if (ei
->type
!= E820_RAM
||
206 ei
->addr
+ei
->size
<= start
||
210 addr
= round_up(ei
->addr
, PAGE_SIZE
);
214 last
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
221 return ((end
- start
) - ram
) >> PAGE_SHIFT
;
225 * Mark e820 reserved areas as busy for the resource manager.
227 void __init
e820_reserve_resources(void)
230 for (i
= 0; i
< e820
.nr_map
; i
++) {
231 struct resource
*res
;
232 res
= alloc_bootmem_low(sizeof(struct resource
));
233 switch (e820
.map
[i
].type
) {
234 case E820_RAM
: res
->name
= "System RAM"; break;
235 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
236 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
237 default: res
->name
= "reserved";
239 res
->start
= e820
.map
[i
].addr
;
240 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
241 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
242 request_resource(&iomem_resource
, res
);
243 if (e820
.map
[i
].type
== E820_RAM
) {
245 * We don't know which RAM region contains kernel data,
246 * so we try it repeatedly and let the resource manager
249 request_resource(res
, &code_resource
);
250 request_resource(res
, &data_resource
);
252 request_resource(res
, &crashk_res
);
259 * Add a memory region to the kernel e820 map.
261 void __init
add_memory_region(unsigned long start
, unsigned long size
, int type
)
266 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
270 e820
.map
[x
].addr
= start
;
271 e820
.map
[x
].size
= size
;
272 e820
.map
[x
].type
= type
;
276 void __init
e820_print_map(char *who
)
280 for (i
= 0; i
< e820
.nr_map
; i
++) {
281 printk(" %s: %016Lx - %016Lx ", who
,
282 (unsigned long long) e820
.map
[i
].addr
,
283 (unsigned long long) (e820
.map
[i
].addr
+ e820
.map
[i
].size
));
284 switch (e820
.map
[i
].type
) {
285 case E820_RAM
: printk("(usable)\n");
288 printk("(reserved)\n");
291 printk("(ACPI data)\n");
294 printk("(ACPI NVS)\n");
296 default: printk("type %u\n", e820
.map
[i
].type
);
303 * Sanitize the BIOS e820 map.
305 * Some e820 responses include overlapping entries. The following
306 * replaces the original e820 map with a new one, removing overlaps.
309 static int __init
sanitize_e820_map(struct e820entry
* biosmap
, char * pnr_map
)
311 struct change_member
{
312 struct e820entry
*pbios
; /* pointer to original bios entry */
313 unsigned long long addr
; /* address for this change point */
315 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
316 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
317 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
318 static struct e820entry new_bios
[E820MAX
] __initdata
;
319 struct change_member
*change_tmp
;
320 unsigned long current_type
, last_type
;
321 unsigned long long last_addr
;
322 int chgidx
, still_changing
;
325 int old_nr
, new_nr
, chg_nr
;
329 Visually we're performing the following (1,2,3,4 = memory types)...
331 Sample memory map (w/overlaps):
332 ____22__________________
333 ______________________4_
334 ____1111________________
335 _44_____________________
336 11111111________________
337 ____________________33__
338 ___________44___________
339 __________33333_________
340 ______________22________
341 ___________________2222_
342 _________111111111______
343 _____________________11_
344 _________________4______
346 Sanitized equivalent (no overlap):
347 1_______________________
348 _44_____________________
349 ___1____________________
350 ____22__________________
351 ______11________________
352 _________1______________
353 __________3_____________
354 ___________44___________
355 _____________33_________
356 _______________2________
357 ________________1_______
358 _________________4______
359 ___________________2____
360 ____________________33__
361 ______________________4_
364 /* if there's only one memory region, don't bother */
370 /* bail out if we find any unreasonable addresses in bios map */
371 for (i
=0; i
<old_nr
; i
++)
372 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
)
375 /* create pointers for initial change-point information (for sorting) */
376 for (i
=0; i
< 2*old_nr
; i
++)
377 change_point
[i
] = &change_point_list
[i
];
379 /* record all known change-points (starting and ending addresses),
380 omitting those that are for empty memory regions */
382 for (i
=0; i
< old_nr
; i
++) {
383 if (biosmap
[i
].size
!= 0) {
384 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
385 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
386 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+ biosmap
[i
].size
;
387 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
392 /* sort change-point list by memory addresses (low -> high) */
394 while (still_changing
) {
396 for (i
=1; i
< chg_nr
; i
++) {
397 /* if <current_addr> > <last_addr>, swap */
398 /* or, if current=<start_addr> & last=<end_addr>, swap */
399 if ((change_point
[i
]->addr
< change_point
[i
-1]->addr
) ||
400 ((change_point
[i
]->addr
== change_point
[i
-1]->addr
) &&
401 (change_point
[i
]->addr
== change_point
[i
]->pbios
->addr
) &&
402 (change_point
[i
-1]->addr
!= change_point
[i
-1]->pbios
->addr
))
405 change_tmp
= change_point
[i
];
406 change_point
[i
] = change_point
[i
-1];
407 change_point
[i
-1] = change_tmp
;
413 /* create a new bios memory map, removing overlaps */
414 overlap_entries
=0; /* number of entries in the overlap table */
415 new_bios_entry
=0; /* index for creating new bios map entries */
416 last_type
= 0; /* start with undefined memory type */
417 last_addr
= 0; /* start with 0 as last starting address */
418 /* loop through change-points, determining affect on the new bios map */
419 for (chgidx
=0; chgidx
< chg_nr
; chgidx
++)
421 /* keep track of all overlapping bios entries */
422 if (change_point
[chgidx
]->addr
== change_point
[chgidx
]->pbios
->addr
)
424 /* add map entry to overlap list (> 1 entry implies an overlap) */
425 overlap_list
[overlap_entries
++]=change_point
[chgidx
]->pbios
;
429 /* remove entry from list (order independent, so swap with last) */
430 for (i
=0; i
<overlap_entries
; i
++)
432 if (overlap_list
[i
] == change_point
[chgidx
]->pbios
)
433 overlap_list
[i
] = overlap_list
[overlap_entries
-1];
437 /* if there are overlapping entries, decide which "type" to use */
438 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
440 for (i
=0; i
<overlap_entries
; i
++)
441 if (overlap_list
[i
]->type
> current_type
)
442 current_type
= overlap_list
[i
]->type
;
443 /* continue building up new bios map based on this information */
444 if (current_type
!= last_type
) {
445 if (last_type
!= 0) {
446 new_bios
[new_bios_entry
].size
=
447 change_point
[chgidx
]->addr
- last_addr
;
448 /* move forward only if the new size was non-zero */
449 if (new_bios
[new_bios_entry
].size
!= 0)
450 if (++new_bios_entry
>= E820MAX
)
451 break; /* no more space left for new bios entries */
453 if (current_type
!= 0) {
454 new_bios
[new_bios_entry
].addr
= change_point
[chgidx
]->addr
;
455 new_bios
[new_bios_entry
].type
= current_type
;
456 last_addr
=change_point
[chgidx
]->addr
;
458 last_type
= current_type
;
461 new_nr
= new_bios_entry
; /* retain count for new bios entries */
463 /* copy new bios mapping into original location */
464 memcpy(biosmap
, new_bios
, new_nr
*sizeof(struct e820entry
));
471 * Copy the BIOS e820 map into a safe place.
473 * Sanity-check it while we're at it..
475 * If we're lucky and live on a modern system, the setup code
476 * will have given us a memory map that we can use to properly
477 * set up memory. If we aren't, we'll fake a memory map.
479 * We check to see that the memory map contains at least 2 elements
480 * before we'll use it, because the detection code in setup.S may
481 * not be perfect and most every PC known to man has two memory
482 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
483 * thinkpad 560x, for example, does not cooperate with the memory
486 static int __init
copy_e820_map(struct e820entry
* biosmap
, int nr_map
)
488 /* Only one memory region (or negative)? Ignore it */
493 unsigned long start
= biosmap
->addr
;
494 unsigned long size
= biosmap
->size
;
495 unsigned long end
= start
+ size
;
496 unsigned long type
= biosmap
->type
;
498 /* Overflow in 64 bits? Ignore the memory map. */
503 * Some BIOSes claim RAM in the 640k - 1M region.
504 * Not right. Fix it up.
506 * This should be removed on Hammer which is supposed to not
507 * have non e820 covered ISA mappings there, but I had some strange
508 * problems so it stays for now. -AK
510 if (type
== E820_RAM
) {
511 if (start
< 0x100000ULL
&& end
> 0xA0000ULL
) {
512 if (start
< 0xA0000ULL
)
513 add_memory_region(start
, 0xA0000ULL
-start
, type
);
514 if (end
<= 0x100000ULL
)
521 add_memory_region(start
, size
, type
);
522 } while (biosmap
++,--nr_map
);
526 void __init
setup_memory_region(void)
528 char *who
= "BIOS-e820";
531 * Try to copy the BIOS-supplied E820-map.
533 * Otherwise fake a memory map; one section from 0k->640k,
534 * the next section from 1mb->appropriate_mem_k
536 sanitize_e820_map(E820_MAP
, &E820_MAP_NR
);
537 if (copy_e820_map(E820_MAP
, E820_MAP_NR
) < 0) {
538 unsigned long mem_size
;
540 /* compare results from other methods and take the greater */
541 if (ALT_MEM_K
< EXT_MEM_K
) {
542 mem_size
= EXT_MEM_K
;
545 mem_size
= ALT_MEM_K
;
550 add_memory_region(0, LOWMEMSIZE(), E820_RAM
);
551 add_memory_region(HIGH_MEMORY
, mem_size
<< 10, E820_RAM
);
553 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
557 void __init
parse_memopt(char *p
, char **from
)
559 end_user_pfn
= memparse(p
, from
);
560 end_user_pfn
>>= PAGE_SHIFT
;
563 void __init
parse_memmapopt(char *p
, char **from
)
565 unsigned long long start_at
, mem_size
;
567 mem_size
= memparse(p
, from
);
570 start_at
= memparse(p
+1, from
);
571 add_memory_region(start_at
, mem_size
, E820_RAM
);
572 } else if (*p
== '#') {
573 start_at
= memparse(p
+1, from
);
574 add_memory_region(start_at
, mem_size
, E820_ACPI
);
575 } else if (*p
== '$') {
576 start_at
= memparse(p
+1, from
);
577 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
579 end_user_pfn
= (mem_size
>> PAGE_SHIFT
);
584 unsigned long pci_mem_start
= 0xaeedbabe;
587 * Search for the biggest gap in the low 32 bits of the e820
588 * memory space. We pass this space to PCI to assign MMIO resources
589 * for hotplug or unconfigured devices in.
590 * Hopefully the BIOS let enough space left.
592 __init
void e820_setup_gap(void)
594 unsigned long gapstart
, gapsize
, round
;
599 last
= 0x100000000ull
;
600 gapstart
= 0x10000000;
604 unsigned long long start
= e820
.map
[i
].addr
;
605 unsigned long long end
= start
+ e820
.map
[i
].size
;
608 * Since "last" is at most 4GB, we know we'll
609 * fit in 32 bits if this condition is true
612 unsigned long gap
= last
- end
;
625 gapstart
= (end_pfn
<< PAGE_SHIFT
) + 1024*1024;
626 printk(KERN_ERR
"PCI: Warning: Cannot find a gap in the 32bit address range\n"
627 KERN_ERR
"PCI: Unassigned devices with 32bit resource registers may break!\n");
631 * See how much we want to round up: start off with
632 * rounding to the next 1MB area.
635 while ((gapsize
>> 4) > round
)
637 /* Fun with two's complement */
638 pci_mem_start
= (gapstart
+ round
) & -round
;
640 printk(KERN_INFO
"Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
641 pci_mem_start
, gapstart
, gapsize
);