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>
29 * PFN of last memory page.
31 unsigned long end_pfn;
32 EXPORT_SYMBOL(end_pfn);
35 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
36 * The direct mapping extends to end_pfn_map, so that we can directly access
37 * apertures, ACPI and other tables without having to play with fixmaps.
39 unsigned long end_pfn_map;
42 * Last pfn which the user wants to use.
44 unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT;
46 extern struct resource code_resource, data_resource;
48 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
49 static inline int bad_addr(unsigned long *addrp, unsigned long size)
51 unsigned long addr = *addrp, last = addr + size;
53 /* various gunk below that needed for SMP startup */
59 /* direct mapping tables of the kernel */
60 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
61 *addrp = table_end << PAGE_SHIFT;
66 #ifdef CONFIG_BLK_DEV_INITRD
67 if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
68 addr < INITRD_START+INITRD_SIZE) {
69 *addrp = INITRD_START + INITRD_SIZE;
73 /* kernel code + 640k memory hole (later should not be needed, but
74 be paranoid for now) */
75 if (last >= 640*1024 && addr < 1024*1024) {
79 if (last >= __pa_symbol(&_text) && last < __pa_symbol(&_end)) {
80 *addrp = __pa_symbol(&_end);
84 if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
85 *addrp = ebda_addr + ebda_size;
89 /* XXX ramdisk image here? */
94 * This function checks if any part of the range <start,end> is mapped
98 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
101 for (i = 0; i < e820.nr_map; i++) {
102 struct e820entry *ei = &e820.map[i];
103 if (type && ei->type != type)
105 if (ei->addr >= end || ei->addr + ei->size <= start)
113 * This function checks if the entire range <start,end> is mapped with type.
115 * Note: this function only works correct if the e820 table is sorted and
116 * not-overlapping, which is the case
118 int __init e820_all_mapped(unsigned long start, unsigned long end, unsigned type)
121 for (i = 0; i < e820.nr_map; i++) {
122 struct e820entry *ei = &e820.map[i];
123 if (type && ei->type != type)
125 /* is the region (part) in overlap with the current region ?*/
126 if (ei->addr >= end || ei->addr + ei->size <= start)
129 /* if the region is at the beginning of <start,end> we move
130 * start to the end of the region since it's ok until there
132 if (ei->addr <= start)
133 start = ei->addr + ei->size;
134 /* if start is now at or beyond end, we're done, full coverage */
136 return 1; /* we're done */
142 * Find a free area in a specific range.
144 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
147 for (i = 0; i < e820.nr_map; i++) {
148 struct e820entry *ei = &e820.map[i];
149 unsigned long addr = ei->addr, last;
150 if (ei->type != E820_RAM)
154 if (addr > ei->addr + ei->size)
156 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
159 if (last > ei->addr + ei->size)
169 * Free bootmem based on the e820 table for a node.
171 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
174 for (i = 0; i < e820.nr_map; i++) {
175 struct e820entry *ei = &e820.map[i];
176 unsigned long last, addr;
178 if (ei->type != E820_RAM ||
179 ei->addr+ei->size <= start ||
183 addr = round_up(ei->addr, PAGE_SIZE);
187 last = round_down(ei->addr + ei->size, PAGE_SIZE);
191 if (last > addr && last-addr >= PAGE_SIZE)
192 free_bootmem_node(pgdat, addr, last-addr);
197 * Find the highest page frame number we have available
199 unsigned long __init e820_end_of_ram(void)
202 unsigned long end_pfn = 0;
204 for (i = 0; i < e820.nr_map; i++) {
205 struct e820entry *ei = &e820.map[i];
206 unsigned long start, end;
208 start = round_up(ei->addr, PAGE_SIZE);
209 end = round_down(ei->addr + ei->size, PAGE_SIZE);
212 if (ei->type == E820_RAM) {
213 if (end > end_pfn<<PAGE_SHIFT)
214 end_pfn = end>>PAGE_SHIFT;
216 if (end > end_pfn_map<<PAGE_SHIFT)
217 end_pfn_map = end>>PAGE_SHIFT;
221 if (end_pfn > end_pfn_map)
222 end_pfn_map = end_pfn;
223 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
224 end_pfn_map = MAXMEM>>PAGE_SHIFT;
225 if (end_pfn > end_user_pfn)
226 end_pfn = end_user_pfn;
227 if (end_pfn > end_pfn_map)
228 end_pfn = end_pfn_map;
234 * Compute how much memory is missing in a range.
235 * Unlike the other functions in this file the arguments are in page numbers.
238 e820_hole_size(unsigned long start_pfn, unsigned long end_pfn)
240 unsigned long ram = 0;
241 unsigned long start = start_pfn << PAGE_SHIFT;
242 unsigned long end = end_pfn << PAGE_SHIFT;
244 for (i = 0; i < e820.nr_map; i++) {
245 struct e820entry *ei = &e820.map[i];
246 unsigned long last, addr;
248 if (ei->type != E820_RAM ||
249 ei->addr+ei->size <= start ||
253 addr = round_up(ei->addr, PAGE_SIZE);
257 last = round_down(ei->addr + ei->size, PAGE_SIZE);
264 return ((end - start) - ram) >> PAGE_SHIFT;
268 * Mark e820 reserved areas as busy for the resource manager.
270 void __init e820_reserve_resources(void)
273 for (i = 0; i < e820.nr_map; i++) {
274 struct resource *res;
275 res = alloc_bootmem_low(sizeof(struct resource));
276 switch (e820.map[i].type) {
277 case E820_RAM: res->name = "System RAM"; break;
278 case E820_ACPI: res->name = "ACPI Tables"; break;
279 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
280 default: res->name = "reserved";
282 res->start = e820.map[i].addr;
283 res->end = res->start + e820.map[i].size - 1;
284 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
285 request_resource(&iomem_resource, res);
286 if (e820.map[i].type == E820_RAM) {
288 * We don't know which RAM region contains kernel data,
289 * so we try it repeatedly and let the resource manager
292 request_resource(res, &code_resource);
293 request_resource(res, &data_resource);
295 request_resource(res, &crashk_res);
301 /* Mark pages corresponding to given address range as nosave */
303 e820_mark_nosave_range(unsigned long start, unsigned long end)
305 unsigned long pfn, max_pfn;
310 printk("Nosave address range: %016lx - %016lx\n", start, end);
311 max_pfn = end >> PAGE_SHIFT;
312 for (pfn = start >> PAGE_SHIFT; pfn < max_pfn; pfn++)
314 SetPageNosave(pfn_to_page(pfn));
318 * Find the ranges of physical addresses that do not correspond to
319 * e820 RAM areas and mark the corresponding pages as nosave for software
320 * suspend and suspend to RAM.
322 * This function requires the e820 map to be sorted and without any
323 * overlapping entries and assumes the first e820 area to be RAM.
325 void __init e820_mark_nosave_regions(void)
330 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
331 for (i = 1; i < e820.nr_map; i++) {
332 struct e820entry *ei = &e820.map[i];
334 if (paddr < ei->addr)
335 e820_mark_nosave_range(paddr,
336 round_up(ei->addr, PAGE_SIZE));
338 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
339 if (ei->type != E820_RAM)
340 e820_mark_nosave_range(round_up(ei->addr, PAGE_SIZE),
343 if (paddr >= (end_pfn << PAGE_SHIFT))
349 * Add a memory region to the kernel e820 map.
351 void __init add_memory_region(unsigned long start, unsigned long size, int type)
356 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
360 e820.map[x].addr = start;
361 e820.map[x].size = size;
362 e820.map[x].type = type;
366 void __init e820_print_map(char *who)
370 for (i = 0; i < e820.nr_map; i++) {
371 printk(" %s: %016Lx - %016Lx ", who,
372 (unsigned long long) e820.map[i].addr,
373 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
374 switch (e820.map[i].type) {
375 case E820_RAM: printk("(usable)\n");
378 printk("(reserved)\n");
381 printk("(ACPI data)\n");
384 printk("(ACPI NVS)\n");
386 default: printk("type %u\n", e820.map[i].type);
393 * Sanitize the BIOS e820 map.
395 * Some e820 responses include overlapping entries. The following
396 * replaces the original e820 map with a new one, removing overlaps.
399 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
401 struct change_member {
402 struct e820entry *pbios; /* pointer to original bios entry */
403 unsigned long long addr; /* address for this change point */
405 static struct change_member change_point_list[2*E820MAX] __initdata;
406 static struct change_member *change_point[2*E820MAX] __initdata;
407 static struct e820entry *overlap_list[E820MAX] __initdata;
408 static struct e820entry new_bios[E820MAX] __initdata;
409 struct change_member *change_tmp;
410 unsigned long current_type, last_type;
411 unsigned long long last_addr;
412 int chgidx, still_changing;
415 int old_nr, new_nr, chg_nr;
419 Visually we're performing the following (1,2,3,4 = memory types)...
421 Sample memory map (w/overlaps):
422 ____22__________________
423 ______________________4_
424 ____1111________________
425 _44_____________________
426 11111111________________
427 ____________________33__
428 ___________44___________
429 __________33333_________
430 ______________22________
431 ___________________2222_
432 _________111111111______
433 _____________________11_
434 _________________4______
436 Sanitized equivalent (no overlap):
437 1_______________________
438 _44_____________________
439 ___1____________________
440 ____22__________________
441 ______11________________
442 _________1______________
443 __________3_____________
444 ___________44___________
445 _____________33_________
446 _______________2________
447 ________________1_______
448 _________________4______
449 ___________________2____
450 ____________________33__
451 ______________________4_
454 /* if there's only one memory region, don't bother */
460 /* bail out if we find any unreasonable addresses in bios map */
461 for (i=0; i<old_nr; i++)
462 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
465 /* create pointers for initial change-point information (for sorting) */
466 for (i=0; i < 2*old_nr; i++)
467 change_point[i] = &change_point_list[i];
469 /* record all known change-points (starting and ending addresses),
470 omitting those that are for empty memory regions */
472 for (i=0; i < old_nr; i++) {
473 if (biosmap[i].size != 0) {
474 change_point[chgidx]->addr = biosmap[i].addr;
475 change_point[chgidx++]->pbios = &biosmap[i];
476 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
477 change_point[chgidx++]->pbios = &biosmap[i];
482 /* sort change-point list by memory addresses (low -> high) */
484 while (still_changing) {
486 for (i=1; i < chg_nr; i++) {
487 /* if <current_addr> > <last_addr>, swap */
488 /* or, if current=<start_addr> & last=<end_addr>, swap */
489 if ((change_point[i]->addr < change_point[i-1]->addr) ||
490 ((change_point[i]->addr == change_point[i-1]->addr) &&
491 (change_point[i]->addr == change_point[i]->pbios->addr) &&
492 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
495 change_tmp = change_point[i];
496 change_point[i] = change_point[i-1];
497 change_point[i-1] = change_tmp;
503 /* create a new bios memory map, removing overlaps */
504 overlap_entries=0; /* number of entries in the overlap table */
505 new_bios_entry=0; /* index for creating new bios map entries */
506 last_type = 0; /* start with undefined memory type */
507 last_addr = 0; /* start with 0 as last starting address */
508 /* loop through change-points, determining affect on the new bios map */
509 for (chgidx=0; chgidx < chg_nr; chgidx++)
511 /* keep track of all overlapping bios entries */
512 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
514 /* add map entry to overlap list (> 1 entry implies an overlap) */
515 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
519 /* remove entry from list (order independent, so swap with last) */
520 for (i=0; i<overlap_entries; i++)
522 if (overlap_list[i] == change_point[chgidx]->pbios)
523 overlap_list[i] = overlap_list[overlap_entries-1];
527 /* if there are overlapping entries, decide which "type" to use */
528 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
530 for (i=0; i<overlap_entries; i++)
531 if (overlap_list[i]->type > current_type)
532 current_type = overlap_list[i]->type;
533 /* continue building up new bios map based on this information */
534 if (current_type != last_type) {
535 if (last_type != 0) {
536 new_bios[new_bios_entry].size =
537 change_point[chgidx]->addr - last_addr;
538 /* move forward only if the new size was non-zero */
539 if (new_bios[new_bios_entry].size != 0)
540 if (++new_bios_entry >= E820MAX)
541 break; /* no more space left for new bios entries */
543 if (current_type != 0) {
544 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
545 new_bios[new_bios_entry].type = current_type;
546 last_addr=change_point[chgidx]->addr;
548 last_type = current_type;
551 new_nr = new_bios_entry; /* retain count for new bios entries */
553 /* copy new bios mapping into original location */
554 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
561 * Copy the BIOS e820 map into a safe place.
563 * Sanity-check it while we're at it..
565 * If we're lucky and live on a modern system, the setup code
566 * will have given us a memory map that we can use to properly
567 * set up memory. If we aren't, we'll fake a memory map.
569 * We check to see that the memory map contains at least 2 elements
570 * before we'll use it, because the detection code in setup.S may
571 * not be perfect and most every PC known to man has two memory
572 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
573 * thinkpad 560x, for example, does not cooperate with the memory
576 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
578 /* Only one memory region (or negative)? Ignore it */
583 unsigned long start = biosmap->addr;
584 unsigned long size = biosmap->size;
585 unsigned long end = start + size;
586 unsigned long type = biosmap->type;
588 /* Overflow in 64 bits? Ignore the memory map. */
593 * Some BIOSes claim RAM in the 640k - 1M region.
594 * Not right. Fix it up.
596 * This should be removed on Hammer which is supposed to not
597 * have non e820 covered ISA mappings there, but I had some strange
598 * problems so it stays for now. -AK
600 if (type == E820_RAM) {
601 if (start < 0x100000ULL && end > 0xA0000ULL) {
602 if (start < 0xA0000ULL)
603 add_memory_region(start, 0xA0000ULL-start, type);
604 if (end <= 0x100000ULL)
611 add_memory_region(start, size, type);
612 } while (biosmap++,--nr_map);
616 void __init setup_memory_region(void)
618 char *who = "BIOS-e820";
621 * Try to copy the BIOS-supplied E820-map.
623 * Otherwise fake a memory map; one section from 0k->640k,
624 * the next section from 1mb->appropriate_mem_k
626 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
627 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
628 unsigned long mem_size;
630 /* compare results from other methods and take the greater */
631 if (ALT_MEM_K < EXT_MEM_K) {
632 mem_size = EXT_MEM_K;
635 mem_size = ALT_MEM_K;
640 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
641 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
643 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
647 void __init parse_memopt(char *p, char **from)
649 end_user_pfn = memparse(p, from);
650 end_user_pfn >>= PAGE_SHIFT;
653 void __init parse_memmapopt(char *p, char **from)
655 unsigned long long start_at, mem_size;
657 mem_size = memparse(p, from);
660 start_at = memparse(p+1, from);
661 add_memory_region(start_at, mem_size, E820_RAM);
662 } else if (*p == '#') {
663 start_at = memparse(p+1, from);
664 add_memory_region(start_at, mem_size, E820_ACPI);
665 } else if (*p == '$') {
666 start_at = memparse(p+1, from);
667 add_memory_region(start_at, mem_size, E820_RESERVED);
669 end_user_pfn = (mem_size >> PAGE_SHIFT);
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);