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>
21 #include <asm/pgtable.h>
24 #include <asm/proto.h>
25 #include <asm/bootsetup.h>
26 #include <asm/sections.h>
31 * PFN of last memory page.
33 unsigned long end_pfn
;
34 EXPORT_SYMBOL(end_pfn
);
37 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
38 * The direct mapping extends to end_pfn_map, so that we can directly access
39 * apertures, ACPI and other tables without having to play with fixmaps.
41 unsigned long end_pfn_map
;
44 * Last pfn which the user wants to use.
46 static unsigned long __initdata end_user_pfn
= MAXMEM
>>PAGE_SHIFT
;
48 extern struct resource code_resource
, data_resource
;
50 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
51 static inline int bad_addr(unsigned long *addrp
, unsigned long size
)
53 unsigned long addr
= *addrp
, last
= addr
+ size
;
55 /* various gunk below that needed for SMP startup */
57 *addrp
= PAGE_ALIGN(0x8000);
61 /* direct mapping tables of the kernel */
62 if (last
>= table_start
<<PAGE_SHIFT
&& addr
< table_end
<<PAGE_SHIFT
) {
63 *addrp
= PAGE_ALIGN(table_end
<< PAGE_SHIFT
);
68 #ifdef CONFIG_BLK_DEV_INITRD
69 if (LOADER_TYPE
&& INITRD_START
&& last
>= INITRD_START
&&
70 addr
< INITRD_START
+INITRD_SIZE
) {
71 *addrp
= PAGE_ALIGN(INITRD_START
+ INITRD_SIZE
);
76 if (last
>= __pa_symbol(&_text
) && addr
< __pa_symbol(&_end
)) {
77 *addrp
= PAGE_ALIGN(__pa_symbol(&_end
));
81 if (last
>= ebda_addr
&& addr
< ebda_addr
+ ebda_size
) {
82 *addrp
= PAGE_ALIGN(ebda_addr
+ ebda_size
);
87 /* NUMA memory to node map */
88 if (last
>= nodemap_addr
&& addr
< nodemap_addr
+ nodemap_size
) {
89 *addrp
= nodemap_addr
+ nodemap_size
;
93 /* XXX ramdisk image here? */
98 * This function checks if any part of the range <start,end> is mapped
102 e820_any_mapped(unsigned long start
, unsigned long end
, unsigned type
)
105 for (i
= 0; i
< e820
.nr_map
; i
++) {
106 struct e820entry
*ei
= &e820
.map
[i
];
107 if (type
&& ei
->type
!= type
)
109 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
115 EXPORT_SYMBOL_GPL(e820_any_mapped
);
118 * This function checks if the entire range <start,end> is mapped with type.
120 * Note: this function only works correct if the e820 table is sorted and
121 * not-overlapping, which is the case
123 int __init
e820_all_mapped(unsigned long start
, unsigned long end
, unsigned type
)
126 for (i
= 0; i
< e820
.nr_map
; i
++) {
127 struct e820entry
*ei
= &e820
.map
[i
];
128 if (type
&& ei
->type
!= type
)
130 /* is the region (part) in overlap with the current region ?*/
131 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
134 /* if the region is at the beginning of <start,end> we move
135 * start to the end of the region since it's ok until there
137 if (ei
->addr
<= start
)
138 start
= ei
->addr
+ ei
->size
;
139 /* if start is now at or beyond end, we're done, full coverage */
141 return 1; /* we're done */
147 * Find a free area in a specific range.
149 unsigned long __init
find_e820_area(unsigned long start
, unsigned long end
, unsigned size
)
152 for (i
= 0; i
< e820
.nr_map
; i
++) {
153 struct e820entry
*ei
= &e820
.map
[i
];
154 unsigned long addr
= ei
->addr
, last
;
155 if (ei
->type
!= E820_RAM
)
159 if (addr
> ei
->addr
+ ei
->size
)
161 while (bad_addr(&addr
, size
) && addr
+size
<= ei
->addr
+ei
->size
)
163 last
= PAGE_ALIGN(addr
) + size
;
164 if (last
> ei
->addr
+ ei
->size
)
174 * Find the highest page frame number we have available
176 unsigned long __init
e820_end_of_ram(void)
178 unsigned long end_pfn
= 0;
179 end_pfn
= find_max_pfn_with_active_regions();
181 if (end_pfn
> end_pfn_map
)
182 end_pfn_map
= end_pfn
;
183 if (end_pfn_map
> MAXMEM
>>PAGE_SHIFT
)
184 end_pfn_map
= MAXMEM
>>PAGE_SHIFT
;
185 if (end_pfn
> end_user_pfn
)
186 end_pfn
= end_user_pfn
;
187 if (end_pfn
> end_pfn_map
)
188 end_pfn
= end_pfn_map
;
190 printk("end_pfn_map = %lu\n", end_pfn_map
);
195 * Find the hole size in the range.
197 unsigned long __init
e820_hole_size(unsigned long start
, unsigned long end
)
199 unsigned long ram
= 0;
202 for (i
= 0; i
< e820
.nr_map
; i
++) {
203 struct e820entry
*ei
= &e820
.map
[i
];
204 unsigned long last
, addr
;
206 if (ei
->type
!= E820_RAM
||
207 ei
->addr
+ei
->size
<= start
||
211 addr
= round_up(ei
->addr
, PAGE_SIZE
);
215 last
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
222 return ((end
- start
) - ram
);
226 * Mark e820 reserved areas as busy for the resource manager.
228 void __init
e820_reserve_resources(void)
231 for (i
= 0; i
< e820
.nr_map
; i
++) {
232 struct resource
*res
;
233 res
= alloc_bootmem_low(sizeof(struct resource
));
234 switch (e820
.map
[i
].type
) {
235 case E820_RAM
: res
->name
= "System RAM"; break;
236 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
237 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
238 default: res
->name
= "reserved";
240 res
->start
= e820
.map
[i
].addr
;
241 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
242 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
243 request_resource(&iomem_resource
, res
);
244 if (e820
.map
[i
].type
== E820_RAM
) {
246 * We don't know which RAM region contains kernel data,
247 * so we try it repeatedly and let the resource manager
250 request_resource(res
, &code_resource
);
251 request_resource(res
, &data_resource
);
253 request_resource(res
, &crashk_res
);
259 /* Mark pages corresponding to given address range as nosave */
261 e820_mark_nosave_range(unsigned long start
, unsigned long end
)
263 unsigned long pfn
, max_pfn
;
268 printk("Nosave address range: %016lx - %016lx\n", start
, end
);
269 max_pfn
= end
>> PAGE_SHIFT
;
270 for (pfn
= start
>> PAGE_SHIFT
; pfn
< max_pfn
; pfn
++)
272 SetPageNosave(pfn_to_page(pfn
));
276 * Find the ranges of physical addresses that do not correspond to
277 * e820 RAM areas and mark the corresponding pages as nosave for software
278 * suspend and suspend to RAM.
280 * This function requires the e820 map to be sorted and without any
281 * overlapping entries and assumes the first e820 area to be RAM.
283 void __init
e820_mark_nosave_regions(void)
288 paddr
= round_down(e820
.map
[0].addr
+ e820
.map
[0].size
, PAGE_SIZE
);
289 for (i
= 1; i
< e820
.nr_map
; i
++) {
290 struct e820entry
*ei
= &e820
.map
[i
];
292 if (paddr
< ei
->addr
)
293 e820_mark_nosave_range(paddr
,
294 round_up(ei
->addr
, PAGE_SIZE
));
296 paddr
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
);
297 if (ei
->type
!= E820_RAM
)
298 e820_mark_nosave_range(round_up(ei
->addr
, PAGE_SIZE
),
301 if (paddr
>= (end_pfn
<< PAGE_SHIFT
))
306 /* Walk the e820 map and register active regions within a node */
308 e820_register_active_regions(int nid
, unsigned long start_pfn
,
309 unsigned long end_pfn
)
312 unsigned long ei_startpfn
, ei_endpfn
;
313 for (i
= 0; i
< e820
.nr_map
; i
++) {
314 struct e820entry
*ei
= &e820
.map
[i
];
315 ei_startpfn
= round_up(ei
->addr
, PAGE_SIZE
) >> PAGE_SHIFT
;
316 ei_endpfn
= round_down(ei
->addr
+ ei
->size
, PAGE_SIZE
)
319 /* Skip map entries smaller than a page */
320 if (ei_startpfn
>= ei_endpfn
)
323 /* Check if end_pfn_map should be updated */
324 if (ei
->type
!= E820_RAM
&& ei_endpfn
> end_pfn_map
)
325 end_pfn_map
= ei_endpfn
;
327 /* Skip if map is outside the node */
328 if (ei
->type
!= E820_RAM
||
329 ei_endpfn
<= start_pfn
||
330 ei_startpfn
>= end_pfn
)
333 /* Check for overlaps */
334 if (ei_startpfn
< start_pfn
)
335 ei_startpfn
= start_pfn
;
336 if (ei_endpfn
> end_pfn
)
339 /* Obey end_user_pfn to save on memmap */
340 if (ei_startpfn
>= end_user_pfn
)
342 if (ei_endpfn
> end_user_pfn
)
343 ei_endpfn
= end_user_pfn
;
345 add_active_range(nid
, ei_startpfn
, ei_endpfn
);
350 * Add a memory region to the kernel e820 map.
352 void __init
add_memory_region(unsigned long start
, unsigned long size
, int type
)
357 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
361 e820
.map
[x
].addr
= start
;
362 e820
.map
[x
].size
= size
;
363 e820
.map
[x
].type
= type
;
367 void __init
e820_print_map(char *who
)
371 for (i
= 0; i
< e820
.nr_map
; i
++) {
372 printk(" %s: %016Lx - %016Lx ", who
,
373 (unsigned long long) e820
.map
[i
].addr
,
374 (unsigned long long) (e820
.map
[i
].addr
+ e820
.map
[i
].size
));
375 switch (e820
.map
[i
].type
) {
376 case E820_RAM
: printk("(usable)\n");
379 printk("(reserved)\n");
382 printk("(ACPI data)\n");
385 printk("(ACPI NVS)\n");
387 default: printk("type %u\n", e820
.map
[i
].type
);
394 * Sanitize the BIOS e820 map.
396 * Some e820 responses include overlapping entries. The following
397 * replaces the original e820 map with a new one, removing overlaps.
400 static int __init
sanitize_e820_map(struct e820entry
* biosmap
, char * pnr_map
)
402 struct change_member
{
403 struct e820entry
*pbios
; /* pointer to original bios entry */
404 unsigned long long addr
; /* address for this change point */
406 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
407 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
408 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
409 static struct e820entry new_bios
[E820MAX
] __initdata
;
410 struct change_member
*change_tmp
;
411 unsigned long current_type
, last_type
;
412 unsigned long long last_addr
;
413 int chgidx
, still_changing
;
416 int old_nr
, new_nr
, chg_nr
;
420 Visually we're performing the following (1,2,3,4 = memory types)...
422 Sample memory map (w/overlaps):
423 ____22__________________
424 ______________________4_
425 ____1111________________
426 _44_____________________
427 11111111________________
428 ____________________33__
429 ___________44___________
430 __________33333_________
431 ______________22________
432 ___________________2222_
433 _________111111111______
434 _____________________11_
435 _________________4______
437 Sanitized equivalent (no overlap):
438 1_______________________
439 _44_____________________
440 ___1____________________
441 ____22__________________
442 ______11________________
443 _________1______________
444 __________3_____________
445 ___________44___________
446 _____________33_________
447 _______________2________
448 ________________1_______
449 _________________4______
450 ___________________2____
451 ____________________33__
452 ______________________4_
455 /* if there's only one memory region, don't bother */
461 /* bail out if we find any unreasonable addresses in bios map */
462 for (i
=0; i
<old_nr
; i
++)
463 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
)
466 /* create pointers for initial change-point information (for sorting) */
467 for (i
=0; i
< 2*old_nr
; i
++)
468 change_point
[i
] = &change_point_list
[i
];
470 /* record all known change-points (starting and ending addresses),
471 omitting those that are for empty memory regions */
473 for (i
=0; i
< old_nr
; i
++) {
474 if (biosmap
[i
].size
!= 0) {
475 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
476 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
477 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+ biosmap
[i
].size
;
478 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
483 /* sort change-point list by memory addresses (low -> high) */
485 while (still_changing
) {
487 for (i
=1; i
< chg_nr
; i
++) {
488 /* if <current_addr> > <last_addr>, swap */
489 /* or, if current=<start_addr> & last=<end_addr>, swap */
490 if ((change_point
[i
]->addr
< change_point
[i
-1]->addr
) ||
491 ((change_point
[i
]->addr
== change_point
[i
-1]->addr
) &&
492 (change_point
[i
]->addr
== change_point
[i
]->pbios
->addr
) &&
493 (change_point
[i
-1]->addr
!= change_point
[i
-1]->pbios
->addr
))
496 change_tmp
= change_point
[i
];
497 change_point
[i
] = change_point
[i
-1];
498 change_point
[i
-1] = change_tmp
;
504 /* create a new bios memory map, removing overlaps */
505 overlap_entries
=0; /* number of entries in the overlap table */
506 new_bios_entry
=0; /* index for creating new bios map entries */
507 last_type
= 0; /* start with undefined memory type */
508 last_addr
= 0; /* start with 0 as last starting address */
509 /* loop through change-points, determining affect on the new bios map */
510 for (chgidx
=0; chgidx
< chg_nr
; chgidx
++)
512 /* keep track of all overlapping bios entries */
513 if (change_point
[chgidx
]->addr
== change_point
[chgidx
]->pbios
->addr
)
515 /* add map entry to overlap list (> 1 entry implies an overlap) */
516 overlap_list
[overlap_entries
++]=change_point
[chgidx
]->pbios
;
520 /* remove entry from list (order independent, so swap with last) */
521 for (i
=0; i
<overlap_entries
; i
++)
523 if (overlap_list
[i
] == change_point
[chgidx
]->pbios
)
524 overlap_list
[i
] = overlap_list
[overlap_entries
-1];
528 /* if there are overlapping entries, decide which "type" to use */
529 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
531 for (i
=0; i
<overlap_entries
; i
++)
532 if (overlap_list
[i
]->type
> current_type
)
533 current_type
= overlap_list
[i
]->type
;
534 /* continue building up new bios map based on this information */
535 if (current_type
!= last_type
) {
536 if (last_type
!= 0) {
537 new_bios
[new_bios_entry
].size
=
538 change_point
[chgidx
]->addr
- last_addr
;
539 /* move forward only if the new size was non-zero */
540 if (new_bios
[new_bios_entry
].size
!= 0)
541 if (++new_bios_entry
>= E820MAX
)
542 break; /* no more space left for new bios entries */
544 if (current_type
!= 0) {
545 new_bios
[new_bios_entry
].addr
= change_point
[chgidx
]->addr
;
546 new_bios
[new_bios_entry
].type
= current_type
;
547 last_addr
=change_point
[chgidx
]->addr
;
549 last_type
= current_type
;
552 new_nr
= new_bios_entry
; /* retain count for new bios entries */
554 /* copy new bios mapping into original location */
555 memcpy(biosmap
, new_bios
, new_nr
*sizeof(struct e820entry
));
562 * Copy the BIOS e820 map into a safe place.
564 * Sanity-check it while we're at it..
566 * If we're lucky and live on a modern system, the setup code
567 * will have given us a memory map that we can use to properly
568 * set up memory. If we aren't, we'll fake a memory map.
570 static int __init
copy_e820_map(struct e820entry
* biosmap
, int nr_map
)
572 /* Only one memory region (or negative)? Ignore it */
577 unsigned long start
= biosmap
->addr
;
578 unsigned long size
= biosmap
->size
;
579 unsigned long end
= start
+ size
;
580 unsigned long type
= biosmap
->type
;
582 /* Overflow in 64 bits? Ignore the memory map. */
586 add_memory_region(start
, size
, type
);
587 } while (biosmap
++,--nr_map
);
591 void early_panic(char *msg
)
597 void __init
setup_memory_region(void)
600 * Try to copy the BIOS-supplied E820-map.
602 * Otherwise fake a memory map; one section from 0k->640k,
603 * the next section from 1mb->appropriate_mem_k
605 sanitize_e820_map(E820_MAP
, &E820_MAP_NR
);
606 if (copy_e820_map(E820_MAP
, E820_MAP_NR
) < 0)
607 early_panic("Cannot find a valid memory map");
608 printk(KERN_INFO
"BIOS-provided physical RAM map:\n");
609 e820_print_map("BIOS-e820");
612 static int __init
parse_memopt(char *p
)
616 end_user_pfn
= memparse(p
, &p
);
617 end_user_pfn
>>= PAGE_SHIFT
;
620 early_param("mem", parse_memopt
);
622 static int userdef __initdata
;
624 static int __init
parse_memmap_opt(char *p
)
627 unsigned long long start_at
, mem_size
;
629 if (!strcmp(p
, "exactmap")) {
630 #ifdef CONFIG_CRASH_DUMP
631 /* If we are doing a crash dump, we
632 * still need to know the real mem
633 * size before original memory map is
636 e820_register_active_regions(0, 0, -1UL);
637 saved_max_pfn
= e820_end_of_ram();
638 remove_all_active_ranges();
647 mem_size
= memparse(p
, &p
);
651 start_at
= memparse(p
+1, &p
);
652 add_memory_region(start_at
, mem_size
, E820_RAM
);
653 } else if (*p
== '#') {
654 start_at
= memparse(p
+1, &p
);
655 add_memory_region(start_at
, mem_size
, E820_ACPI
);
656 } else if (*p
== '$') {
657 start_at
= memparse(p
+1, &p
);
658 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
660 end_user_pfn
= (mem_size
>> PAGE_SHIFT
);
662 return *p
== '\0' ? 0 : -EINVAL
;
664 early_param("memmap", parse_memmap_opt
);
666 void __init
finish_e820_parsing(void)
669 printk(KERN_INFO
"user-defined physical RAM map:\n");
670 e820_print_map("user");
674 unsigned long pci_mem_start
= 0xaeedbabe;
675 EXPORT_SYMBOL(pci_mem_start
);
678 * Search for the biggest gap in the low 32 bits of the e820
679 * memory space. We pass this space to PCI to assign MMIO resources
680 * for hotplug or unconfigured devices in.
681 * Hopefully the BIOS let enough space left.
683 __init
void e820_setup_gap(void)
685 unsigned long gapstart
, gapsize
, round
;
690 last
= 0x100000000ull
;
691 gapstart
= 0x10000000;
695 unsigned long long start
= e820
.map
[i
].addr
;
696 unsigned long long end
= start
+ e820
.map
[i
].size
;
699 * Since "last" is at most 4GB, we know we'll
700 * fit in 32 bits if this condition is true
703 unsigned long gap
= last
- end
;
716 gapstart
= (end_pfn
<< PAGE_SHIFT
) + 1024*1024;
717 printk(KERN_ERR
"PCI: Warning: Cannot find a gap in the 32bit address range\n"
718 KERN_ERR
"PCI: Unassigned devices with 32bit resource registers may break!\n");
722 * See how much we want to round up: start off with
723 * rounding to the next 1MB area.
726 while ((gapsize
>> 4) > round
)
728 /* Fun with two's complement */
729 pci_mem_start
= (gapstart
+ round
) & -round
;
731 printk(KERN_INFO
"Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
732 pci_mem_start
, gapstart
, gapsize
);