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 <asm/pgtable.h>
23 #include <asm/proto.h>
24 #include <asm/bootsetup.h>
25 #include <asm/sections.h>
27 struct e820map e820 __initdata
;
30 * PFN of last memory page.
32 unsigned long end_pfn
;
33 EXPORT_SYMBOL(end_pfn
);
36 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
37 * The direct mapping extends to end_pfn_map, so that we can directly access
38 * apertures, ACPI and other tables without having to play with fixmaps.
40 unsigned long end_pfn_map
;
43 * Last pfn which the user wants to use.
45 static unsigned long __initdata end_user_pfn
= MAXMEM
>>PAGE_SHIFT
;
47 extern struct resource code_resource
, data_resource
;
49 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
50 static inline int bad_addr(unsigned long *addrp
, unsigned long size
)
52 unsigned long addr
= *addrp
, last
= addr
+ size
;
54 /* various gunk below that needed for SMP startup */
60 /* direct mapping tables of the kernel */
61 if (last
>= table_start
<<PAGE_SHIFT
&& addr
< table_end
<<PAGE_SHIFT
) {
62 *addrp
= table_end
<< PAGE_SHIFT
;
67 #ifdef CONFIG_BLK_DEV_INITRD
68 if (LOADER_TYPE
&& INITRD_START
&& last
>= INITRD_START
&&
69 addr
< INITRD_START
+INITRD_SIZE
) {
70 *addrp
= INITRD_START
+ INITRD_SIZE
;
75 if (last
>= __pa_symbol(&_text
) && last
< __pa_symbol(&_end
)) {
76 *addrp
= __pa_symbol(&_end
);
80 if (last
>= ebda_addr
&& addr
< ebda_addr
+ ebda_size
) {
81 *addrp
= ebda_addr
+ ebda_size
;
85 /* XXX ramdisk image here? */
90 * This function checks if any part of the range <start,end> is mapped
94 e820_any_mapped(unsigned long start
, unsigned long end
, unsigned type
)
97 for (i
= 0; i
< e820
.nr_map
; i
++) {
98 struct e820entry
*ei
= &e820
.map
[i
];
99 if (type
&& ei
->type
!= type
)
101 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
109 * This function checks if the entire range <start,end> is mapped with type.
111 * Note: this function only works correct if the e820 table is sorted and
112 * not-overlapping, which is the case
114 int __init
e820_all_mapped(unsigned long start
, unsigned long end
, unsigned type
)
117 for (i
= 0; i
< e820
.nr_map
; i
++) {
118 struct e820entry
*ei
= &e820
.map
[i
];
119 if (type
&& ei
->type
!= type
)
121 /* is the region (part) in overlap with the current region ?*/
122 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
125 /* if the region is at the beginning of <start,end> we move
126 * start to the end of the region since it's ok until there
128 if (ei
->addr
<= start
)
129 start
= ei
->addr
+ ei
->size
;
130 /* if start is now at or beyond end, we're done, full coverage */
132 return 1; /* we're done */
138 * Find a free area in a specific range.
140 unsigned long __init
find_e820_area(unsigned long start
, unsigned long end
, unsigned size
)
143 for (i
= 0; i
< e820
.nr_map
; i
++) {
144 struct e820entry
*ei
= &e820
.map
[i
];
145 unsigned long addr
= ei
->addr
, last
;
146 if (ei
->type
!= E820_RAM
)
150 if (addr
> ei
->addr
+ ei
->size
)
152 while (bad_addr(&addr
, size
) && addr
+size
<= ei
->addr
+ei
->size
)
155 if (last
> ei
->addr
+ ei
->size
)
165 * Free bootmem based on the e820 table for a node.
167 void __init
e820_bootmem_free(pg_data_t
*pgdat
, unsigned long start
,unsigned long end
)
170 for (i
= 0; i
< e820
.nr_map
; i
++) {
171 struct e820entry
*ei
= &e820
.map
[i
];
172 unsigned long last
, addr
;
174 if (ei
->type
!= E820_RAM
||
175 ei
->addr
+ei
->size
<= start
||
179 addr
= round_up(ei
->addr
, PAGE_SIZE
);
183 last
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
187 if (last
> addr
&& last
-addr
>= PAGE_SIZE
)
188 free_bootmem_node(pgdat
, addr
, last
-addr
);
193 * Find the highest page frame number we have available
195 unsigned long __init
e820_end_of_ram(void)
198 unsigned long end_pfn
= 0;
200 for (i
= 0; i
< e820
.nr_map
; i
++) {
201 struct e820entry
*ei
= &e820
.map
[i
];
202 unsigned long start
, end
;
204 start
= round_up(ei
->addr
, PAGE_SIZE
);
205 end
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
208 if (ei
->type
== E820_RAM
) {
209 if (end
> end_pfn
<<PAGE_SHIFT
)
210 end_pfn
= end
>>PAGE_SHIFT
;
212 if (end
> end_pfn_map
<<PAGE_SHIFT
)
213 end_pfn_map
= end
>>PAGE_SHIFT
;
217 if (end_pfn
> end_pfn_map
)
218 end_pfn_map
= end_pfn
;
219 if (end_pfn_map
> MAXMEM
>>PAGE_SHIFT
)
220 end_pfn_map
= MAXMEM
>>PAGE_SHIFT
;
221 if (end_pfn
> end_user_pfn
)
222 end_pfn
= end_user_pfn
;
223 if (end_pfn
> end_pfn_map
)
224 end_pfn
= end_pfn_map
;
230 * Compute how much memory is missing in a range.
231 * Unlike the other functions in this file the arguments are in page numbers.
234 e820_hole_size(unsigned long start_pfn
, unsigned long end_pfn
)
236 unsigned long ram
= 0;
237 unsigned long start
= start_pfn
<< PAGE_SHIFT
;
238 unsigned long end
= end_pfn
<< PAGE_SHIFT
;
240 for (i
= 0; i
< e820
.nr_map
; i
++) {
241 struct e820entry
*ei
= &e820
.map
[i
];
242 unsigned long last
, addr
;
244 if (ei
->type
!= E820_RAM
||
245 ei
->addr
+ei
->size
<= start
||
249 addr
= round_up(ei
->addr
, PAGE_SIZE
);
253 last
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
260 return ((end
- start
) - ram
) >> PAGE_SHIFT
;
264 * Mark e820 reserved areas as busy for the resource manager.
266 void __init
e820_reserve_resources(void)
269 for (i
= 0; i
< e820
.nr_map
; i
++) {
270 struct resource
*res
;
271 res
= alloc_bootmem_low(sizeof(struct resource
));
272 switch (e820
.map
[i
].type
) {
273 case E820_RAM
: res
->name
= "System RAM"; break;
274 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
275 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
276 default: res
->name
= "reserved";
278 res
->start
= e820
.map
[i
].addr
;
279 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
280 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
281 request_resource(&iomem_resource
, res
);
282 if (e820
.map
[i
].type
== E820_RAM
) {
284 * We don't know which RAM region contains kernel data,
285 * so we try it repeatedly and let the resource manager
288 request_resource(res
, &code_resource
);
289 request_resource(res
, &data_resource
);
291 request_resource(res
, &crashk_res
);
298 * Add a memory region to the kernel e820 map.
300 void __init
add_memory_region(unsigned long start
, unsigned long size
, int type
)
305 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
309 e820
.map
[x
].addr
= start
;
310 e820
.map
[x
].size
= size
;
311 e820
.map
[x
].type
= type
;
315 void __init
e820_print_map(char *who
)
319 for (i
= 0; i
< e820
.nr_map
; i
++) {
320 printk(" %s: %016Lx - %016Lx ", who
,
321 (unsigned long long) e820
.map
[i
].addr
,
322 (unsigned long long) (e820
.map
[i
].addr
+ e820
.map
[i
].size
));
323 switch (e820
.map
[i
].type
) {
324 case E820_RAM
: printk("(usable)\n");
327 printk("(reserved)\n");
330 printk("(ACPI data)\n");
333 printk("(ACPI NVS)\n");
335 default: printk("type %u\n", e820
.map
[i
].type
);
342 * Sanitize the BIOS e820 map.
344 * Some e820 responses include overlapping entries. The following
345 * replaces the original e820 map with a new one, removing overlaps.
348 static int __init
sanitize_e820_map(struct e820entry
* biosmap
, char * pnr_map
)
350 struct change_member
{
351 struct e820entry
*pbios
; /* pointer to original bios entry */
352 unsigned long long addr
; /* address for this change point */
354 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
355 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
356 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
357 static struct e820entry new_bios
[E820MAX
] __initdata
;
358 struct change_member
*change_tmp
;
359 unsigned long current_type
, last_type
;
360 unsigned long long last_addr
;
361 int chgidx
, still_changing
;
364 int old_nr
, new_nr
, chg_nr
;
368 Visually we're performing the following (1,2,3,4 = memory types)...
370 Sample memory map (w/overlaps):
371 ____22__________________
372 ______________________4_
373 ____1111________________
374 _44_____________________
375 11111111________________
376 ____________________33__
377 ___________44___________
378 __________33333_________
379 ______________22________
380 ___________________2222_
381 _________111111111______
382 _____________________11_
383 _________________4______
385 Sanitized equivalent (no overlap):
386 1_______________________
387 _44_____________________
388 ___1____________________
389 ____22__________________
390 ______11________________
391 _________1______________
392 __________3_____________
393 ___________44___________
394 _____________33_________
395 _______________2________
396 ________________1_______
397 _________________4______
398 ___________________2____
399 ____________________33__
400 ______________________4_
403 /* if there's only one memory region, don't bother */
409 /* bail out if we find any unreasonable addresses in bios map */
410 for (i
=0; i
<old_nr
; i
++)
411 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
)
414 /* create pointers for initial change-point information (for sorting) */
415 for (i
=0; i
< 2*old_nr
; i
++)
416 change_point
[i
] = &change_point_list
[i
];
418 /* record all known change-points (starting and ending addresses),
419 omitting those that are for empty memory regions */
421 for (i
=0; i
< old_nr
; i
++) {
422 if (biosmap
[i
].size
!= 0) {
423 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
424 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
425 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+ biosmap
[i
].size
;
426 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
431 /* sort change-point list by memory addresses (low -> high) */
433 while (still_changing
) {
435 for (i
=1; i
< chg_nr
; i
++) {
436 /* if <current_addr> > <last_addr>, swap */
437 /* or, if current=<start_addr> & last=<end_addr>, swap */
438 if ((change_point
[i
]->addr
< change_point
[i
-1]->addr
) ||
439 ((change_point
[i
]->addr
== change_point
[i
-1]->addr
) &&
440 (change_point
[i
]->addr
== change_point
[i
]->pbios
->addr
) &&
441 (change_point
[i
-1]->addr
!= change_point
[i
-1]->pbios
->addr
))
444 change_tmp
= change_point
[i
];
445 change_point
[i
] = change_point
[i
-1];
446 change_point
[i
-1] = change_tmp
;
452 /* create a new bios memory map, removing overlaps */
453 overlap_entries
=0; /* number of entries in the overlap table */
454 new_bios_entry
=0; /* index for creating new bios map entries */
455 last_type
= 0; /* start with undefined memory type */
456 last_addr
= 0; /* start with 0 as last starting address */
457 /* loop through change-points, determining affect on the new bios map */
458 for (chgidx
=0; chgidx
< chg_nr
; chgidx
++)
460 /* keep track of all overlapping bios entries */
461 if (change_point
[chgidx
]->addr
== change_point
[chgidx
]->pbios
->addr
)
463 /* add map entry to overlap list (> 1 entry implies an overlap) */
464 overlap_list
[overlap_entries
++]=change_point
[chgidx
]->pbios
;
468 /* remove entry from list (order independent, so swap with last) */
469 for (i
=0; i
<overlap_entries
; i
++)
471 if (overlap_list
[i
] == change_point
[chgidx
]->pbios
)
472 overlap_list
[i
] = overlap_list
[overlap_entries
-1];
476 /* if there are overlapping entries, decide which "type" to use */
477 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
479 for (i
=0; i
<overlap_entries
; i
++)
480 if (overlap_list
[i
]->type
> current_type
)
481 current_type
= overlap_list
[i
]->type
;
482 /* continue building up new bios map based on this information */
483 if (current_type
!= last_type
) {
484 if (last_type
!= 0) {
485 new_bios
[new_bios_entry
].size
=
486 change_point
[chgidx
]->addr
- last_addr
;
487 /* move forward only if the new size was non-zero */
488 if (new_bios
[new_bios_entry
].size
!= 0)
489 if (++new_bios_entry
>= E820MAX
)
490 break; /* no more space left for new bios entries */
492 if (current_type
!= 0) {
493 new_bios
[new_bios_entry
].addr
= change_point
[chgidx
]->addr
;
494 new_bios
[new_bios_entry
].type
= current_type
;
495 last_addr
=change_point
[chgidx
]->addr
;
497 last_type
= current_type
;
500 new_nr
= new_bios_entry
; /* retain count for new bios entries */
502 /* copy new bios mapping into original location */
503 memcpy(biosmap
, new_bios
, new_nr
*sizeof(struct e820entry
));
510 * Copy the BIOS e820 map into a safe place.
512 * Sanity-check it while we're at it..
514 * If we're lucky and live on a modern system, the setup code
515 * will have given us a memory map that we can use to properly
516 * set up memory. If we aren't, we'll fake a memory map.
518 static int __init
copy_e820_map(struct e820entry
* biosmap
, int nr_map
)
520 /* Only one memory region (or negative)? Ignore it */
525 unsigned long start
= biosmap
->addr
;
526 unsigned long size
= biosmap
->size
;
527 unsigned long end
= start
+ size
;
528 unsigned long type
= biosmap
->type
;
530 /* Overflow in 64 bits? Ignore the memory map. */
534 add_memory_region(start
, size
, type
);
535 } while (biosmap
++,--nr_map
);
539 void __init
setup_memory_region(void)
541 char *who
= "BIOS-e820";
544 * Try to copy the BIOS-supplied E820-map.
546 * Otherwise fake a memory map; one section from 0k->640k,
547 * the next section from 1mb->appropriate_mem_k
549 sanitize_e820_map(E820_MAP
, &E820_MAP_NR
);
550 if (copy_e820_map(E820_MAP
, E820_MAP_NR
) < 0) {
551 unsigned long mem_size
;
553 /* compare results from other methods and take the greater */
554 if (ALT_MEM_K
< EXT_MEM_K
) {
555 mem_size
= EXT_MEM_K
;
558 mem_size
= ALT_MEM_K
;
563 add_memory_region(0, LOWMEMSIZE(), E820_RAM
);
564 add_memory_region(HIGH_MEMORY
, mem_size
<< 10, E820_RAM
);
566 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
570 static int __init
parse_memopt(char *p
)
574 end_user_pfn
= memparse(p
, &p
);
575 end_user_pfn
>>= PAGE_SHIFT
;
578 early_param("mem", parse_memopt
);
580 static int userdef __initdata
;
582 static int __init
parse_memmap_opt(char *p
)
585 unsigned long long start_at
, mem_size
;
587 if (!strcmp(p
, "exactmap")) {
588 #ifdef CONFIG_CRASH_DUMP
589 /* If we are doing a crash dump, we
590 * still need to know the real mem
591 * size before original memory map is
594 saved_max_pfn
= e820_end_of_ram();
603 mem_size
= memparse(p
, &p
);
607 start_at
= memparse(p
+1, &p
);
608 add_memory_region(start_at
, mem_size
, E820_RAM
);
609 } else if (*p
== '#') {
610 start_at
= memparse(p
+1, &p
);
611 add_memory_region(start_at
, mem_size
, E820_ACPI
);
612 } else if (*p
== '$') {
613 start_at
= memparse(p
+1, &p
);
614 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
616 end_user_pfn
= (mem_size
>> PAGE_SHIFT
);
618 return *p
== '\0' ? 0 : -EINVAL
;
620 early_param("memmap", parse_memmap_opt
);
622 void finish_e820_parsing(void)
625 printk(KERN_INFO
"user-defined physical RAM map:\n");
626 e820_print_map("user");
630 unsigned long pci_mem_start
= 0xaeedbabe;
631 EXPORT_SYMBOL(pci_mem_start
);
634 * Search for the biggest gap in the low 32 bits of the e820
635 * memory space. We pass this space to PCI to assign MMIO resources
636 * for hotplug or unconfigured devices in.
637 * Hopefully the BIOS let enough space left.
639 __init
void e820_setup_gap(void)
641 unsigned long gapstart
, gapsize
, round
;
646 last
= 0x100000000ull
;
647 gapstart
= 0x10000000;
651 unsigned long long start
= e820
.map
[i
].addr
;
652 unsigned long long end
= start
+ e820
.map
[i
].size
;
655 * Since "last" is at most 4GB, we know we'll
656 * fit in 32 bits if this condition is true
659 unsigned long gap
= last
- end
;
672 gapstart
= (end_pfn
<< PAGE_SHIFT
) + 1024*1024;
673 printk(KERN_ERR
"PCI: Warning: Cannot find a gap in the 32bit address range\n"
674 KERN_ERR
"PCI: Unassigned devices with 32bit resource registers may break!\n");
678 * See how much we want to round up: start off with
679 * rounding to the next 1MB area.
682 while ((gapsize
>> 4) > round
)
684 /* Fun with two's complement */
685 pci_mem_start
= (gapstart
+ round
) & -round
;
687 printk(KERN_INFO
"Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
688 pci_mem_start
, gapstart
, gapsize
);