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)
322 size_t total_size = 0;
324 for (i = 0; i < e820.nr_map; i++) {
325 const struct e820entry * entry = &e820.map[i];
326 switch (entry->type) {
327 case E820_RAM: sprintf(type, "(usable)\n");
328 total_size += entry->size;
331 sprintf(type, "(reserved)\n");
334 sprintf(type, "(ACPI data)\n");
337 sprintf(type, "(ACPI NVS)\n");
339 default: sprintf(type, "type %u\n", entry->type);
344 " %s: %016Lx - %016Lx %s",
346 (unsigned long long) entry->addr,
347 (unsigned long long) (entry->addr + entry->size),
351 printk( KERN_DEBUG "Total usable memory %ld bytes\n", total_size );
355 * Sanitize the BIOS e820 map.
357 * Some e820 responses include overlapping entries. The following
358 * replaces the original e820 map with a new one, removing overlaps.
361 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
363 struct change_member {
364 struct e820entry *pbios; /* pointer to original bios entry */
365 unsigned long long addr; /* address for this change point */
367 static struct change_member change_point_list[2*E820MAX] __initdata;
368 static struct change_member *change_point[2*E820MAX] __initdata;
369 static struct e820entry *overlap_list[E820MAX] __initdata;
370 static struct e820entry new_bios[E820MAX] __initdata;
371 struct change_member *change_tmp;
372 unsigned long current_type, last_type;
373 unsigned long long last_addr;
374 int chgidx, still_changing;
377 int old_nr, new_nr, chg_nr;
381 Visually we're performing the following (1,2,3,4 = memory types)...
383 Sample memory map (w/overlaps):
384 ____22__________________
385 ______________________4_
386 ____1111________________
387 _44_____________________
388 11111111________________
389 ____________________33__
390 ___________44___________
391 __________33333_________
392 ______________22________
393 ___________________2222_
394 _________111111111______
395 _____________________11_
396 _________________4______
398 Sanitized equivalent (no overlap):
399 1_______________________
400 _44_____________________
401 ___1____________________
402 ____22__________________
403 ______11________________
404 _________1______________
405 __________3_____________
406 ___________44___________
407 _____________33_________
408 _______________2________
409 ________________1_______
410 _________________4______
411 ___________________2____
412 ____________________33__
413 ______________________4_
416 /* if there's only one memory region, don't bother */
422 /* bail out if we find any unreasonable addresses in bios map */
423 for (i=0; i<old_nr; i++)
424 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
427 /* create pointers for initial change-point information (for sorting) */
428 for (i=0; i < 2*old_nr; i++)
429 change_point[i] = &change_point_list[i];
431 /* record all known change-points (starting and ending addresses),
432 omitting those that are for empty memory regions */
434 for (i=0; i < old_nr; i++) {
435 if (biosmap[i].size != 0) {
436 change_point[chgidx]->addr = biosmap[i].addr;
437 change_point[chgidx++]->pbios = &biosmap[i];
438 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
439 change_point[chgidx++]->pbios = &biosmap[i];
444 /* sort change-point list by memory addresses (low -> high) */
446 while (still_changing) {
448 for (i=1; i < chg_nr; i++) {
449 /* if <current_addr> > <last_addr>, swap */
450 /* or, if current=<start_addr> & last=<end_addr>, swap */
451 if ((change_point[i]->addr < change_point[i-1]->addr) ||
452 ((change_point[i]->addr == change_point[i-1]->addr) &&
453 (change_point[i]->addr == change_point[i]->pbios->addr) &&
454 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
457 change_tmp = change_point[i];
458 change_point[i] = change_point[i-1];
459 change_point[i-1] = change_tmp;
465 /* create a new bios memory map, removing overlaps */
466 overlap_entries=0; /* number of entries in the overlap table */
467 new_bios_entry=0; /* index for creating new bios map entries */
468 last_type = 0; /* start with undefined memory type */
469 last_addr = 0; /* start with 0 as last starting address */
470 /* loop through change-points, determining affect on the new bios map */
471 for (chgidx=0; chgidx < chg_nr; chgidx++)
473 /* keep track of all overlapping bios entries */
474 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
476 /* add map entry to overlap list (> 1 entry implies an overlap) */
477 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
481 /* remove entry from list (order independent, so swap with last) */
482 for (i=0; i<overlap_entries; i++)
484 if (overlap_list[i] == change_point[chgidx]->pbios)
485 overlap_list[i] = overlap_list[overlap_entries-1];
489 /* if there are overlapping entries, decide which "type" to use */
490 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
492 for (i=0; i<overlap_entries; i++)
493 if (overlap_list[i]->type > current_type)
494 current_type = overlap_list[i]->type;
495 /* continue building up new bios map based on this information */
496 if (current_type != last_type) {
497 if (last_type != 0) {
498 new_bios[new_bios_entry].size =
499 change_point[chgidx]->addr - last_addr;
500 /* move forward only if the new size was non-zero */
501 if (new_bios[new_bios_entry].size != 0)
502 if (++new_bios_entry >= E820MAX)
503 break; /* no more space left for new bios entries */
505 if (current_type != 0) {
506 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
507 new_bios[new_bios_entry].type = current_type;
508 last_addr=change_point[chgidx]->addr;
510 last_type = current_type;
513 new_nr = new_bios_entry; /* retain count for new bios entries */
515 /* copy new bios mapping into original location */
516 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
523 * Copy the BIOS e820 map into a safe place.
525 * Sanity-check it while we're at it..
527 * If we're lucky and live on a modern system, the setup code
528 * will have given us a memory map that we can use to properly
529 * set up memory. If we aren't, we'll fake a memory map.
531 * We check to see that the memory map contains at least 2 elements
532 * before we'll use it, because the detection code in setup.S may
533 * not be perfect and most every PC known to man has two memory
534 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
535 * thinkpad 560x, for example, does not cooperate with the memory
538 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
540 /* Only one memory region (or negative)? Ignore it */
545 unsigned long start = biosmap->addr;
546 unsigned long size = biosmap->size;
547 unsigned long end = start + size;
548 unsigned long type = biosmap->type;
550 /* Overflow in 64 bits? Ignore the memory map. */
555 * Some BIOSes claim RAM in the 640k - 1M region.
556 * Not right. Fix it up.
558 * This should be removed on Hammer which is supposed to not
559 * have non e820 covered ISA mappings there, but I had some strange
560 * problems so it stays for now. -AK
562 if (type == E820_RAM) {
563 if (start < 0x100000ULL && end > 0xA0000ULL) {
564 if (start < 0xA0000ULL)
565 add_memory_region(start, 0xA0000ULL-start, type);
566 if (end <= 0x100000ULL)
573 add_memory_region(start, size, type);
574 } while (biosmap++,--nr_map);
578 void __init setup_memory_region(void)
580 char *who = "BIOS-e820";
583 * Try to copy the BIOS-supplied E820-map.
585 * Otherwise fake a memory map; one section from 0k->640k,
586 * the next section from 1mb->appropriate_mem_k
588 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
589 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
590 unsigned long mem_size;
592 /* compare results from other methods and take the greater */
593 if (ALT_MEM_K < EXT_MEM_K) {
594 mem_size = EXT_MEM_K;
597 mem_size = ALT_MEM_K;
602 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
603 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
606 printk(KERN_DEBUG "BIOS-provided physical RAM map:\n");
609 /* This also sets end_pfn_map */
610 end_pfn = e820_end_of_ram();
613 void __init parse_memopt(char *p, char **from)
615 end_user_pfn = memparse(p, from);
616 end_user_pfn >>= PAGE_SHIFT;
619 void __init parse_memmapopt(char *p, char **from)
621 unsigned long long start_at, mem_size;
623 mem_size = memparse(p, from);
626 start_at = memparse(p+1, from);
627 add_memory_region(start_at, mem_size, E820_RAM);
628 } else if (*p == '#') {
629 start_at = memparse(p+1, from);
630 add_memory_region(start_at, mem_size, E820_ACPI);
631 } else if (*p == '$') {
632 start_at = memparse(p+1, from);
633 add_memory_region(start_at, mem_size, E820_RESERVED);
635 end_user_pfn = (mem_size >> PAGE_SHIFT);
640 unsigned long pci_mem_start = 0xaeedbabe;
643 * Search for the biggest gap in the low 32 bits of the e820
644 * memory space. We pass this space to PCI to assign MMIO resources
645 * for hotplug or unconfigured devices in.
646 * Hopefully the BIOS let enough space left.
648 __init void e820_setup_gap(void)
650 unsigned long gapstart, gapsize, round;
655 last = 0x100000000ull;
656 gapstart = 0x10000000;
660 unsigned long long start = e820.map[i].addr;
661 unsigned long long end = start + e820.map[i].size;
664 * Since "last" is at most 4GB, we know we'll
665 * fit in 32 bits if this condition is true
668 unsigned long gap = last - end;
681 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
682 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
683 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
687 * See how much we want to round up: start off with
688 * rounding to the next 1MB area.
691 while ((gapsize >> 4) > round)
693 /* Fun with two's complement */
694 pci_mem_start = (gapstart + round) & -round;
696 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
697 pci_mem_start, gapstart, gapsize);