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/bootsetup.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 (LOADER_TYPE
&& INITRD_START
&& last
>= INITRD_START
&&
72 addr
< INITRD_START
+INITRD_SIZE
) {
73 *addrp
= PAGE_ALIGN(INITRD_START
+ INITRD_SIZE
);
78 if (last
>= __pa_symbol(&_text
) && addr
< __pa_symbol(&_end
)) {
79 *addrp
= PAGE_ALIGN(__pa_symbol(&_end
));
83 if (last
>= ebda_addr
&& addr
< ebda_addr
+ ebda_size
) {
84 *addrp
= PAGE_ALIGN(ebda_addr
+ ebda_size
);
89 /* NUMA memory to node map */
90 if (last
>= nodemap_addr
&& addr
< nodemap_addr
+ nodemap_size
) {
91 *addrp
= nodemap_addr
+ nodemap_size
;
95 /* XXX ramdisk image here? */
100 * This function checks if any part of the range <start,end> is mapped
104 e820_any_mapped(unsigned long start
, unsigned long end
, unsigned type
)
107 for (i
= 0; i
< e820
.nr_map
; i
++) {
108 struct e820entry
*ei
= &e820
.map
[i
];
109 if (type
&& ei
->type
!= type
)
111 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
117 EXPORT_SYMBOL_GPL(e820_any_mapped
);
120 * This function checks if the entire range <start,end> is mapped with type.
122 * Note: this function only works correct if the e820 table is sorted and
123 * not-overlapping, which is the case
125 int __init
e820_all_mapped(unsigned long start
, unsigned long end
, unsigned type
)
128 for (i
= 0; i
< e820
.nr_map
; i
++) {
129 struct e820entry
*ei
= &e820
.map
[i
];
130 if (type
&& ei
->type
!= type
)
132 /* is the region (part) in overlap with the current region ?*/
133 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
136 /* if the region is at the beginning of <start,end> we move
137 * start to the end of the region since it's ok until there
139 if (ei
->addr
<= start
)
140 start
= ei
->addr
+ ei
->size
;
141 /* if start is now at or beyond end, we're done, full coverage */
143 return 1; /* we're done */
149 * Find a free area in a specific range.
151 unsigned long __init
find_e820_area(unsigned long start
, unsigned long end
, unsigned size
)
154 for (i
= 0; i
< e820
.nr_map
; i
++) {
155 struct e820entry
*ei
= &e820
.map
[i
];
156 unsigned long addr
= ei
->addr
, last
;
157 if (ei
->type
!= E820_RAM
)
161 if (addr
> ei
->addr
+ ei
->size
)
163 while (bad_addr(&addr
, size
) && addr
+size
<= ei
->addr
+ei
->size
)
165 last
= PAGE_ALIGN(addr
) + size
;
166 if (last
> ei
->addr
+ ei
->size
)
176 * Find the highest page frame number we have available
178 unsigned long __init
e820_end_of_ram(void)
180 unsigned long end_pfn
= 0;
181 end_pfn
= find_max_pfn_with_active_regions();
183 if (end_pfn
> end_pfn_map
)
184 end_pfn_map
= end_pfn
;
185 if (end_pfn_map
> MAXMEM
>>PAGE_SHIFT
)
186 end_pfn_map
= MAXMEM
>>PAGE_SHIFT
;
187 if (end_pfn
> end_user_pfn
)
188 end_pfn
= end_user_pfn
;
189 if (end_pfn
> end_pfn_map
)
190 end_pfn
= end_pfn_map
;
192 printk("end_pfn_map = %lu\n", end_pfn_map
);
197 * Find the hole size in the range.
199 unsigned long __init
e820_hole_size(unsigned long start
, unsigned long end
)
201 unsigned long ram
= 0;
204 for (i
= 0; i
< e820
.nr_map
; i
++) {
205 struct e820entry
*ei
= &e820
.map
[i
];
206 unsigned long last
, addr
;
208 if (ei
->type
!= E820_RAM
||
209 ei
->addr
+ei
->size
<= start
||
213 addr
= round_up(ei
->addr
, PAGE_SIZE
);
217 last
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
224 return ((end
- start
) - ram
);
228 * Mark e820 reserved areas as busy for the resource manager.
230 void __init
e820_reserve_resources(void)
233 for (i
= 0; i
< e820
.nr_map
; i
++) {
234 struct resource
*res
;
235 res
= alloc_bootmem_low(sizeof(struct resource
));
236 switch (e820
.map
[i
].type
) {
237 case E820_RAM
: res
->name
= "System RAM"; break;
238 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
239 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
240 default: res
->name
= "reserved";
242 res
->start
= e820
.map
[i
].addr
;
243 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
244 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
245 request_resource(&iomem_resource
, res
);
246 if (e820
.map
[i
].type
== E820_RAM
) {
248 * We don't know which RAM region contains kernel data,
249 * so we try it repeatedly and let the resource manager
252 request_resource(res
, &code_resource
);
253 request_resource(res
, &data_resource
);
255 request_resource(res
, &crashk_res
);
262 * Find the ranges of physical addresses that do not correspond to
263 * e820 RAM areas and mark the corresponding pages as nosave for software
264 * suspend and suspend to RAM.
266 * This function requires the e820 map to be sorted and without any
267 * overlapping entries and assumes the first e820 area to be RAM.
269 void __init
e820_mark_nosave_regions(void)
274 paddr
= round_down(e820
.map
[0].addr
+ e820
.map
[0].size
, PAGE_SIZE
);
275 for (i
= 1; i
< e820
.nr_map
; i
++) {
276 struct e820entry
*ei
= &e820
.map
[i
];
278 if (paddr
< ei
->addr
)
279 register_nosave_region(PFN_DOWN(paddr
),
282 paddr
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
283 if (ei
->type
!= E820_RAM
)
284 register_nosave_region(PFN_UP(ei
->addr
),
287 if (paddr
>= (end_pfn
<< PAGE_SHIFT
))
292 /* Walk the e820 map and register active regions within a node */
294 e820_register_active_regions(int nid
, unsigned long start_pfn
,
295 unsigned long end_pfn
)
298 unsigned long ei_startpfn
, ei_endpfn
;
299 for (i
= 0; i
< e820
.nr_map
; i
++) {
300 struct e820entry
*ei
= &e820
.map
[i
];
301 ei_startpfn
= round_up(ei
->addr
, PAGE_SIZE
) >> PAGE_SHIFT
;
302 ei_endpfn
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
)
305 /* Skip map entries smaller than a page */
306 if (ei_startpfn
>= ei_endpfn
)
309 /* Check if end_pfn_map should be updated */
310 if (ei
->type
!= E820_RAM
&& ei_endpfn
> end_pfn_map
)
311 end_pfn_map
= ei_endpfn
;
313 /* Skip if map is outside the node */
314 if (ei
->type
!= E820_RAM
||
315 ei_endpfn
<= start_pfn
||
316 ei_startpfn
>= end_pfn
)
319 /* Check for overlaps */
320 if (ei_startpfn
< start_pfn
)
321 ei_startpfn
= start_pfn
;
322 if (ei_endpfn
> end_pfn
)
325 /* Obey end_user_pfn to save on memmap */
326 if (ei_startpfn
>= end_user_pfn
)
328 if (ei_endpfn
> end_user_pfn
)
329 ei_endpfn
= end_user_pfn
;
331 add_active_range(nid
, ei_startpfn
, ei_endpfn
);
336 * Add a memory region to the kernel e820 map.
338 void __init
add_memory_region(unsigned long start
, unsigned long size
, int type
)
343 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
347 e820
.map
[x
].addr
= start
;
348 e820
.map
[x
].size
= size
;
349 e820
.map
[x
].type
= type
;
353 void __init
e820_print_map(char *who
)
357 for (i
= 0; i
< e820
.nr_map
; i
++) {
358 printk(" %s: %016Lx - %016Lx ", who
,
359 (unsigned long long) e820
.map
[i
].addr
,
360 (unsigned long long) (e820
.map
[i
].addr
+ e820
.map
[i
].size
));
361 switch (e820
.map
[i
].type
) {
362 case E820_RAM
: printk("(usable)\n");
365 printk("(reserved)\n");
368 printk("(ACPI data)\n");
371 printk("(ACPI NVS)\n");
373 default: printk("type %u\n", e820
.map
[i
].type
);
380 * Sanitize the BIOS e820 map.
382 * Some e820 responses include overlapping entries. The following
383 * replaces the original e820 map with a new one, removing overlaps.
386 static int __init
sanitize_e820_map(struct e820entry
* biosmap
, char * pnr_map
)
388 struct change_member
{
389 struct e820entry
*pbios
; /* pointer to original bios entry */
390 unsigned long long addr
; /* address for this change point */
392 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
393 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
394 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
395 static struct e820entry new_bios
[E820MAX
] __initdata
;
396 struct change_member
*change_tmp
;
397 unsigned long current_type
, last_type
;
398 unsigned long long last_addr
;
399 int chgidx
, still_changing
;
402 int old_nr
, new_nr
, chg_nr
;
406 Visually we're performing the following (1,2,3,4 = memory types)...
408 Sample memory map (w/overlaps):
409 ____22__________________
410 ______________________4_
411 ____1111________________
412 _44_____________________
413 11111111________________
414 ____________________33__
415 ___________44___________
416 __________33333_________
417 ______________22________
418 ___________________2222_
419 _________111111111______
420 _____________________11_
421 _________________4______
423 Sanitized equivalent (no overlap):
424 1_______________________
425 _44_____________________
426 ___1____________________
427 ____22__________________
428 ______11________________
429 _________1______________
430 __________3_____________
431 ___________44___________
432 _____________33_________
433 _______________2________
434 ________________1_______
435 _________________4______
436 ___________________2____
437 ____________________33__
438 ______________________4_
441 /* if there's only one memory region, don't bother */
447 /* bail out if we find any unreasonable addresses in bios map */
448 for (i
=0; i
<old_nr
; i
++)
449 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
)
452 /* create pointers for initial change-point information (for sorting) */
453 for (i
=0; i
< 2*old_nr
; i
++)
454 change_point
[i
] = &change_point_list
[i
];
456 /* record all known change-points (starting and ending addresses),
457 omitting those that are for empty memory regions */
459 for (i
=0; i
< old_nr
; i
++) {
460 if (biosmap
[i
].size
!= 0) {
461 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
462 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
463 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+ biosmap
[i
].size
;
464 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
469 /* sort change-point list by memory addresses (low -> high) */
471 while (still_changing
) {
473 for (i
=1; i
< chg_nr
; i
++) {
474 /* if <current_addr> > <last_addr>, swap */
475 /* or, if current=<start_addr> & last=<end_addr>, swap */
476 if ((change_point
[i
]->addr
< change_point
[i
-1]->addr
) ||
477 ((change_point
[i
]->addr
== change_point
[i
-1]->addr
) &&
478 (change_point
[i
]->addr
== change_point
[i
]->pbios
->addr
) &&
479 (change_point
[i
-1]->addr
!= change_point
[i
-1]->pbios
->addr
))
482 change_tmp
= change_point
[i
];
483 change_point
[i
] = change_point
[i
-1];
484 change_point
[i
-1] = change_tmp
;
490 /* create a new bios memory map, removing overlaps */
491 overlap_entries
=0; /* number of entries in the overlap table */
492 new_bios_entry
=0; /* index for creating new bios map entries */
493 last_type
= 0; /* start with undefined memory type */
494 last_addr
= 0; /* start with 0 as last starting address */
495 /* loop through change-points, determining affect on the new bios map */
496 for (chgidx
=0; chgidx
< chg_nr
; chgidx
++)
498 /* keep track of all overlapping bios entries */
499 if (change_point
[chgidx
]->addr
== change_point
[chgidx
]->pbios
->addr
)
501 /* add map entry to overlap list (> 1 entry implies an overlap) */
502 overlap_list
[overlap_entries
++]=change_point
[chgidx
]->pbios
;
506 /* remove entry from list (order independent, so swap with last) */
507 for (i
=0; i
<overlap_entries
; i
++)
509 if (overlap_list
[i
] == change_point
[chgidx
]->pbios
)
510 overlap_list
[i
] = overlap_list
[overlap_entries
-1];
514 /* if there are overlapping entries, decide which "type" to use */
515 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
517 for (i
=0; i
<overlap_entries
; i
++)
518 if (overlap_list
[i
]->type
> current_type
)
519 current_type
= overlap_list
[i
]->type
;
520 /* continue building up new bios map based on this information */
521 if (current_type
!= last_type
) {
522 if (last_type
!= 0) {
523 new_bios
[new_bios_entry
].size
=
524 change_point
[chgidx
]->addr
- last_addr
;
525 /* move forward only if the new size was non-zero */
526 if (new_bios
[new_bios_entry
].size
!= 0)
527 if (++new_bios_entry
>= E820MAX
)
528 break; /* no more space left for new bios entries */
530 if (current_type
!= 0) {
531 new_bios
[new_bios_entry
].addr
= change_point
[chgidx
]->addr
;
532 new_bios
[new_bios_entry
].type
= current_type
;
533 last_addr
=change_point
[chgidx
]->addr
;
535 last_type
= current_type
;
538 new_nr
= new_bios_entry
; /* retain count for new bios entries */
540 /* copy new bios mapping into original location */
541 memcpy(biosmap
, new_bios
, new_nr
*sizeof(struct e820entry
));
548 * Copy the BIOS e820 map into a safe place.
550 * Sanity-check it while we're at it..
552 * If we're lucky and live on a modern system, the setup code
553 * will have given us a memory map that we can use to properly
554 * set up memory. If we aren't, we'll fake a memory map.
556 static int __init
copy_e820_map(struct e820entry
* biosmap
, int nr_map
)
558 /* Only one memory region (or negative)? Ignore it */
563 unsigned long start
= biosmap
->addr
;
564 unsigned long size
= biosmap
->size
;
565 unsigned long end
= start
+ size
;
566 unsigned long type
= biosmap
->type
;
568 /* Overflow in 64 bits? Ignore the memory map. */
572 add_memory_region(start
, size
, type
);
573 } while (biosmap
++,--nr_map
);
577 void early_panic(char *msg
)
583 void __init
setup_memory_region(void)
586 * Try to copy the BIOS-supplied E820-map.
588 * Otherwise fake a memory map; one section from 0k->640k,
589 * the next section from 1mb->appropriate_mem_k
591 sanitize_e820_map(E820_MAP
, &E820_MAP_NR
);
592 if (copy_e820_map(E820_MAP
, E820_MAP_NR
) < 0)
593 early_panic("Cannot find a valid memory map");
594 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
595 e820_print_map("BIOS-e820");
598 static int __init
parse_memopt(char *p
)
602 end_user_pfn
= memparse(p
, &p
);
603 end_user_pfn
>>= PAGE_SHIFT
;
606 early_param("mem", parse_memopt
);
608 static int userdef __initdata
;
610 static int __init
parse_memmap_opt(char *p
)
613 unsigned long long start_at
, mem_size
;
615 if (!strcmp(p
, "exactmap")) {
616 #ifdef CONFIG_CRASH_DUMP
617 /* If we are doing a crash dump, we
618 * still need to know the real mem
619 * size before original memory map is
622 e820_register_active_regions(0, 0, -1UL);
623 saved_max_pfn
= e820_end_of_ram();
624 remove_all_active_ranges();
633 mem_size
= memparse(p
, &p
);
637 start_at
= memparse(p
+1, &p
);
638 add_memory_region(start_at
, mem_size
, E820_RAM
);
639 } else if (*p
== '#') {
640 start_at
= memparse(p
+1, &p
);
641 add_memory_region(start_at
, mem_size
, E820_ACPI
);
642 } else if (*p
== '$') {
643 start_at
= memparse(p
+1, &p
);
644 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
646 end_user_pfn
= (mem_size
>> PAGE_SHIFT
);
648 return *p
== '\0' ? 0 : -EINVAL
;
650 early_param("memmap", parse_memmap_opt
);
652 void __init
finish_e820_parsing(void)
655 printk(KERN_INFO
"user-defined physical RAM map:\n");
656 e820_print_map("user");
660 unsigned long pci_mem_start
= 0xaeedbabe;
661 EXPORT_SYMBOL(pci_mem_start
);
664 * Search for the biggest gap in the low 32 bits of the e820
665 * memory space. We pass this space to PCI to assign MMIO resources
666 * for hotplug or unconfigured devices in.
667 * Hopefully the BIOS let enough space left.
669 __init
void e820_setup_gap(void)
671 unsigned long gapstart
, gapsize
, round
;
676 last
= 0x100000000ull
;
677 gapstart
= 0x10000000;
681 unsigned long long start
= e820
.map
[i
].addr
;
682 unsigned long long end
= start
+ e820
.map
[i
].size
;
685 * Since "last" is at most 4GB, we know we'll
686 * fit in 32 bits if this condition is true
689 unsigned long gap
= last
- end
;
702 gapstart
= (end_pfn
<< PAGE_SHIFT
) + 1024*1024;
703 printk(KERN_ERR
"PCI: Warning: Cannot find a gap in the 32bit address range\n"
704 KERN_ERR
"PCI: Unassigned devices with 32bit resource registers may break!\n");
708 * See how much we want to round up: start off with
709 * rounding to the next 1MB area.
712 while ((gapsize
>> 4) > round
)
714 /* Fun with two's complement */
715 pci_mem_start
= (gapstart
+ round
) & -round
;
717 printk(KERN_INFO
"Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
718 pci_mem_start
, gapstart
, gapsize
);