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 <lwk/kernel.h>
13 #include <lwk/types.h>
15 #include <lwk/bootmem.h>
16 #include <lwk/resource.h>
17 #include <lwk/string.h>
18 #include <lwk/linux_compat.h>
20 #include <arch/page.h>
21 #include <arch/pgtable.h>
22 #include <arch/e820.h>
23 #include <arch/proto.h>
24 #include <arch/bootsetup.h>
25 #include <arch/sections.h>
28 * The BIOS "e820" map of memory.
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 unsigned long 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 */
63 /* direct mapping tables of the kernel */
64 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
65 *addrp = table_end << PAGE_SHIFT;
70 if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
71 addr < INITRD_START+INITRD_SIZE) {
72 *addrp = INITRD_START + INITRD_SIZE;
76 /* kernel code + 640k memory hole (later should not be needed, but
77 be paranoid for now) */
78 if (last >= 640*1024 && addr < __pa_symbol(&_end)) {
79 *addrp = __pa_symbol(&_end);
83 if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
84 *addrp = ebda_addr + ebda_size;
88 /* XXX ramdisk image here? */
93 * This function checks if any part of the range <start,end> is mapped
97 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
100 for (i = 0; i < e820.nr_map; i++) {
101 struct e820entry *ei = &e820.map[i];
102 if (type && ei->type != type)
104 if (ei->addr >= end || ei->addr + ei->size <= start)
112 * This function checks if the entire range <start,end> is mapped with type.
114 * Note: this function only works correct if the e820 table is sorted and
115 * not-overlapping, which is the case
117 int __init e820_all_mapped(unsigned long start, unsigned long end, unsigned type)
120 for (i = 0; i < e820.nr_map; i++) {
121 struct e820entry *ei = &e820.map[i];
122 if (type && ei->type != type)
124 /* is the region (part) in overlap with the current region ?*/
125 if (ei->addr >= end || ei->addr + ei->size <= start)
128 /* if the region is at the beginning of <start,end> we move
129 * start to the end of the region since it's ok until there
131 if (ei->addr <= start)
132 start = ei->addr + ei->size;
133 /* if start is now at or beyond end, we're done, full coverage */
135 return 1; /* we're done */
141 * Find a free area in a specific range.
143 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
146 for (i = 0; i < e820.nr_map; i++) {
147 struct e820entry *ei = &e820.map[i];
148 unsigned long addr = ei->addr, last;
149 if (ei->type != E820_RAM)
153 if (addr > ei->addr + ei->size)
155 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
158 if (last > ei->addr + ei->size)
168 * Free bootmem based on the e820 table for a node.
170 void __init e820_bootmem_free(unsigned long start, unsigned long end)
173 for (i = 0; i < e820.nr_map; i++) {
174 struct e820entry *ei = &e820.map[i];
175 unsigned long last, addr;
177 if (ei->type != E820_RAM ||
178 ei->addr+ei->size <= start ||
182 addr = round_up(ei->addr, PAGE_SIZE);
186 last = round_down(ei->addr + ei->size, PAGE_SIZE);
190 if (last > addr && last-addr >= PAGE_SIZE)
191 free_bootmem(addr, last-addr);
196 * Find the highest page frame number we have available
198 unsigned long __init e820_end_of_ram(void)
201 unsigned long end_pfn = 0;
203 for (i = 0; i < e820.nr_map; i++) {
204 struct e820entry *ei = &e820.map[i];
205 unsigned long start, end;
207 start = round_up(ei->addr, PAGE_SIZE);
208 end = round_down(ei->addr + ei->size, PAGE_SIZE);
211 if (ei->type == E820_RAM) {
212 if (end > end_pfn<<PAGE_SHIFT)
213 end_pfn = end>>PAGE_SHIFT;
215 if (end > end_pfn_map<<PAGE_SHIFT)
216 end_pfn_map = end>>PAGE_SHIFT;
220 if (end_pfn > end_pfn_map)
221 end_pfn_map = end_pfn;
222 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
223 end_pfn_map = MAXMEM>>PAGE_SHIFT;
224 if (end_pfn > end_user_pfn)
225 end_pfn = end_user_pfn;
226 if (end_pfn > end_pfn_map)
227 end_pfn = end_pfn_map;
233 * Compute how much memory is missing in a range.
234 * Unlike the other functions in this file the arguments are in page numbers.
237 e820_hole_size(unsigned long start_pfn, unsigned long end_pfn)
239 unsigned long ram = 0;
240 unsigned long start = start_pfn << PAGE_SHIFT;
241 unsigned long end = end_pfn << PAGE_SHIFT;
243 for (i = 0; i < e820.nr_map; i++) {
244 struct e820entry *ei = &e820.map[i];
245 unsigned long last, addr;
247 if (ei->type != E820_RAM ||
248 ei->addr+ei->size <= start ||
252 addr = round_up(ei->addr, PAGE_SIZE);
256 last = round_down(ei->addr + ei->size, PAGE_SIZE);
263 return ((end - start) - ram) >> PAGE_SHIFT;
267 * Mark e820 reserved areas as busy for the resource manager.
269 void __init e820_reserve_resources(void)
272 for (i = 0; i < e820.nr_map; i++) {
273 struct resource *res;
274 res = alloc_bootmem(sizeof(struct resource));
275 switch (e820.map[i].type) {
276 case E820_RAM: res->name = "System RAM"; break;
277 case E820_ACPI: res->name = "ACPI Tables"; break;
278 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
279 default: res->name = "reserved";
281 res->start = e820.map[i].addr;
282 res->end = res->start + e820.map[i].size - 1;
283 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
284 request_resource(&iomem_resource, res);
285 if (e820.map[i].type == E820_RAM) {
287 * We don't know which RAM region contains kernel data,
288 * so we try it repeatedly and let the resource manager
291 request_resource(res, &code_resource);
292 request_resource(res, &data_resource);
294 request_resource(res, &crashk_res);
301 * Add a memory region to the kernel e820 map.
303 void __init add_memory_region(unsigned long start, unsigned long size, int type)
308 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
312 e820.map[x].addr = start;
313 e820.map[x].size = size;
314 e820.map[x].type = type;
318 void __init e820_print_map(char *who)
323 for (i = 0; i < e820.nr_map; i++) {
324 switch (e820.map[i].type) {
325 case E820_RAM: sprintf(type, "(usable)\n");
328 sprintf(type, "(reserved)\n");
331 sprintf(type, "(ACPI data)\n");
334 sprintf(type, "(ACPI NVS)\n");
336 default: sprintf(type, "type %u\n", e820.map[i].type);
341 " %s: %016Lx - %016Lx %s", who,
342 (unsigned long long) e820.map[i].addr,
343 (unsigned long long) (e820.map[i].addr + e820.map[i].size),
349 * Sanitize the BIOS e820 map.
351 * Some e820 responses include overlapping entries. The following
352 * replaces the original e820 map with a new one, removing overlaps.
355 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
357 struct change_member {
358 struct e820entry *pbios; /* pointer to original bios entry */
359 unsigned long long addr; /* address for this change point */
361 static struct change_member change_point_list[2*E820MAX] __initdata;
362 static struct change_member *change_point[2*E820MAX] __initdata;
363 static struct e820entry *overlap_list[E820MAX] __initdata;
364 static struct e820entry new_bios[E820MAX] __initdata;
365 struct change_member *change_tmp;
366 unsigned long current_type, last_type;
367 unsigned long long last_addr;
368 int chgidx, still_changing;
371 int old_nr, new_nr, chg_nr;
375 Visually we're performing the following (1,2,3,4 = memory types)...
377 Sample memory map (w/overlaps):
378 ____22__________________
379 ______________________4_
380 ____1111________________
381 _44_____________________
382 11111111________________
383 ____________________33__
384 ___________44___________
385 __________33333_________
386 ______________22________
387 ___________________2222_
388 _________111111111______
389 _____________________11_
390 _________________4______
392 Sanitized equivalent (no overlap):
393 1_______________________
394 _44_____________________
395 ___1____________________
396 ____22__________________
397 ______11________________
398 _________1______________
399 __________3_____________
400 ___________44___________
401 _____________33_________
402 _______________2________
403 ________________1_______
404 _________________4______
405 ___________________2____
406 ____________________33__
407 ______________________4_
410 /* if there's only one memory region, don't bother */
416 /* bail out if we find any unreasonable addresses in bios map */
417 for (i=0; i<old_nr; i++)
418 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
421 /* create pointers for initial change-point information (for sorting) */
422 for (i=0; i < 2*old_nr; i++)
423 change_point[i] = &change_point_list[i];
425 /* record all known change-points (starting and ending addresses),
426 omitting those that are for empty memory regions */
428 for (i=0; i < old_nr; i++) {
429 if (biosmap[i].size != 0) {
430 change_point[chgidx]->addr = biosmap[i].addr;
431 change_point[chgidx++]->pbios = &biosmap[i];
432 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
433 change_point[chgidx++]->pbios = &biosmap[i];
438 /* sort change-point list by memory addresses (low -> high) */
440 while (still_changing) {
442 for (i=1; i < chg_nr; i++) {
443 /* if <current_addr> > <last_addr>, swap */
444 /* or, if current=<start_addr> & last=<end_addr>, swap */
445 if ((change_point[i]->addr < change_point[i-1]->addr) ||
446 ((change_point[i]->addr == change_point[i-1]->addr) &&
447 (change_point[i]->addr == change_point[i]->pbios->addr) &&
448 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
451 change_tmp = change_point[i];
452 change_point[i] = change_point[i-1];
453 change_point[i-1] = change_tmp;
459 /* create a new bios memory map, removing overlaps */
460 overlap_entries=0; /* number of entries in the overlap table */
461 new_bios_entry=0; /* index for creating new bios map entries */
462 last_type = 0; /* start with undefined memory type */
463 last_addr = 0; /* start with 0 as last starting address */
464 /* loop through change-points, determining affect on the new bios map */
465 for (chgidx=0; chgidx < chg_nr; chgidx++)
467 /* keep track of all overlapping bios entries */
468 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
470 /* add map entry to overlap list (> 1 entry implies an overlap) */
471 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
475 /* remove entry from list (order independent, so swap with last) */
476 for (i=0; i<overlap_entries; i++)
478 if (overlap_list[i] == change_point[chgidx]->pbios)
479 overlap_list[i] = overlap_list[overlap_entries-1];
483 /* if there are overlapping entries, decide which "type" to use */
484 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
486 for (i=0; i<overlap_entries; i++)
487 if (overlap_list[i]->type > current_type)
488 current_type = overlap_list[i]->type;
489 /* continue building up new bios map based on this information */
490 if (current_type != last_type) {
491 if (last_type != 0) {
492 new_bios[new_bios_entry].size =
493 change_point[chgidx]->addr - last_addr;
494 /* move forward only if the new size was non-zero */
495 if (new_bios[new_bios_entry].size != 0)
496 if (++new_bios_entry >= E820MAX)
497 break; /* no more space left for new bios entries */
499 if (current_type != 0) {
500 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
501 new_bios[new_bios_entry].type = current_type;
502 last_addr=change_point[chgidx]->addr;
504 last_type = current_type;
507 new_nr = new_bios_entry; /* retain count for new bios entries */
509 /* copy new bios mapping into original location */
510 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
517 * Copy the BIOS e820 map into a safe place.
519 * Sanity-check it while we're at it..
521 * If we're lucky and live on a modern system, the setup code
522 * will have given us a memory map that we can use to properly
523 * set up memory. If we aren't, we'll fake a memory map.
525 * We check to see that the memory map contains at least 2 elements
526 * before we'll use it, because the detection code in setup.S may
527 * not be perfect and most every PC known to man has two memory
528 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
529 * thinkpad 560x, for example, does not cooperate with the memory
532 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
534 /* Only one memory region (or negative)? Ignore it */
539 unsigned long start = biosmap->addr;
540 unsigned long size = biosmap->size;
541 unsigned long end = start + size;
542 unsigned long type = biosmap->type;
544 /* Overflow in 64 bits? Ignore the memory map. */
549 * Some BIOSes claim RAM in the 640k - 1M region.
550 * Not right. Fix it up.
552 * This should be removed on Hammer which is supposed to not
553 * have non e820 covered ISA mappings there, but I had some strange
554 * problems so it stays for now. -AK
556 if (type == E820_RAM) {
557 if (start < 0x100000ULL && end > 0xA0000ULL) {
558 if (start < 0xA0000ULL)
559 add_memory_region(start, 0xA0000ULL-start, type);
560 if (end <= 0x100000ULL)
567 add_memory_region(start, size, type);
568 } while (biosmap++,--nr_map);
572 void __init setup_memory_region(void)
574 char *who = "BIOS-e820";
577 * Try to copy the BIOS-supplied E820-map.
579 * Otherwise fake a memory map; one section from 0k->640k,
580 * the next section from 1mb->appropriate_mem_k
582 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
583 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
584 unsigned long mem_size;
586 /* compare results from other methods and take the greater */
587 if (ALT_MEM_K < EXT_MEM_K) {
588 mem_size = EXT_MEM_K;
591 mem_size = ALT_MEM_K;
596 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
597 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
600 printk(KERN_DEBUG "BIOS-provided physical RAM map:\n");
603 /* This also sets end_pfn_map */
604 end_pfn = e820_end_of_ram();
607 void __init parse_memopt(char *p, char **from)
609 end_user_pfn = memparse(p, from);
610 end_user_pfn >>= PAGE_SHIFT;
613 void __init parse_memmapopt(char *p, char **from)
615 unsigned long long start_at, mem_size;
617 mem_size = memparse(p, from);
620 start_at = memparse(p+1, from);
621 add_memory_region(start_at, mem_size, E820_RAM);
622 } else if (*p == '#') {
623 start_at = memparse(p+1, from);
624 add_memory_region(start_at, mem_size, E820_ACPI);
625 } else if (*p == '$') {
626 start_at = memparse(p+1, from);
627 add_memory_region(start_at, mem_size, E820_RESERVED);
629 end_user_pfn = (mem_size >> PAGE_SHIFT);
634 unsigned long pci_mem_start = 0xaeedbabe;
637 * Search for the biggest gap in the low 32 bits of the e820
638 * memory space. We pass this space to PCI to assign MMIO resources
639 * for hotplug or unconfigured devices in.
640 * Hopefully the BIOS let enough space left.
642 __init void e820_setup_gap(void)
644 unsigned long gapstart, gapsize, round;
649 last = 0x100000000ull;
650 gapstart = 0x10000000;
654 unsigned long long start = e820.map[i].addr;
655 unsigned long long end = start + e820.map[i].size;
658 * Since "last" is at most 4GB, we know we'll
659 * fit in 32 bits if this condition is true
662 unsigned long gap = last - end;
675 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
676 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
677 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
681 * See how much we want to round up: start off with
682 * rounding to the next 1MB area.
685 while ((gapsize >> 4) > round)
687 /* Fun with two's complement */
688 pci_mem_start = (gapstart + round) & -round;
690 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
691 pci_mem_start, gapstart, gapsize);