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>
22 #include <asm/proto.h>
23 #include <asm/bootsetup.h>
24 #include <asm/sections.h>
27 * PFN of last memory page.
29 unsigned long end_pfn;
30 EXPORT_SYMBOL(end_pfn);
33 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
34 * The direct mapping extends to end_pfn_map, so that we can directly access
35 * apertures, ACPI and other tables without having to play with fixmaps.
37 unsigned long end_pfn_map;
40 * Last pfn which the user wants to use.
42 unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT;
44 extern struct resource code_resource, data_resource;
46 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
47 static inline int bad_addr(unsigned long *addrp, unsigned long size)
49 unsigned long addr = *addrp, last = addr + size;
51 /* various gunk below that needed for SMP startup */
57 /* direct mapping tables of the kernel */
58 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
59 *addrp = table_end << PAGE_SHIFT;
64 #ifdef CONFIG_BLK_DEV_INITRD
65 if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
66 addr < INITRD_START+INITRD_SIZE) {
67 *addrp = INITRD_START + INITRD_SIZE;
71 /* kernel code + 640k memory hole (later should not be needed, but
72 be paranoid for now) */
73 if (last >= 640*1024 && addr < __pa_symbol(&_end)) {
74 *addrp = __pa_symbol(&_end);
78 if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
79 *addrp = ebda_addr + ebda_size;
83 /* XXX ramdisk image here? */
88 * This function checks if any part of the range <start,end> is mapped
92 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
95 for (i = 0; i < e820.nr_map; i++) {
96 struct e820entry *ei = &e820.map[i];
97 if (type && ei->type != type)
99 if (ei->addr >= end || ei->addr + ei->size <= start)
107 * This function checks if the entire range <start,end> is mapped with type.
109 * Note: this function only works correct if the e820 table is sorted and
110 * not-overlapping, which is the case
112 int __init e820_all_mapped(unsigned long start, unsigned long end, unsigned type)
115 for (i = 0; i < e820.nr_map; i++) {
116 struct e820entry *ei = &e820.map[i];
117 if (type && ei->type != type)
119 /* is the region (part) in overlap with the current region ?*/
120 if (ei->addr >= end || ei->addr + ei->size <= start)
123 /* if the region is at the beginning of <start,end> we move
124 * start to the end of the region since it's ok until there
126 if (ei->addr <= start)
127 start = ei->addr + ei->size;
128 /* if start is now at or beyond end, we're done, full coverage */
130 return 1; /* we're done */
136 * Find a free area in a specific range.
138 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
141 for (i = 0; i < e820.nr_map; i++) {
142 struct e820entry *ei = &e820.map[i];
143 unsigned long addr = ei->addr, last;
144 if (ei->type != E820_RAM)
148 if (addr > ei->addr + ei->size)
150 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
153 if (last > ei->addr + ei->size)
163 * Free bootmem based on the e820 table for a node.
165 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
168 for (i = 0; i < e820.nr_map; i++) {
169 struct e820entry *ei = &e820.map[i];
170 unsigned long last, addr;
172 if (ei->type != E820_RAM ||
173 ei->addr+ei->size <= start ||
177 addr = round_up(ei->addr, PAGE_SIZE);
181 last = round_down(ei->addr + ei->size, PAGE_SIZE);
185 if (last > addr && last-addr >= PAGE_SIZE)
186 free_bootmem_node(pgdat, addr, last-addr);
191 * Find the highest page frame number we have available
193 unsigned long __init e820_end_of_ram(void)
196 unsigned long end_pfn = 0;
198 for (i = 0; i < e820.nr_map; i++) {
199 struct e820entry *ei = &e820.map[i];
200 unsigned long start, end;
202 start = round_up(ei->addr, PAGE_SIZE);
203 end = round_down(ei->addr + ei->size, PAGE_SIZE);
206 if (ei->type == E820_RAM) {
207 if (end > end_pfn<<PAGE_SHIFT)
208 end_pfn = end>>PAGE_SHIFT;
210 if (end > end_pfn_map<<PAGE_SHIFT)
211 end_pfn_map = end>>PAGE_SHIFT;
215 if (end_pfn > end_pfn_map)
216 end_pfn_map = end_pfn;
217 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
218 end_pfn_map = MAXMEM>>PAGE_SHIFT;
219 if (end_pfn > end_user_pfn)
220 end_pfn = end_user_pfn;
221 if (end_pfn > end_pfn_map)
222 end_pfn = end_pfn_map;
228 * Compute how much memory is missing in a range.
229 * Unlike the other functions in this file the arguments are in page numbers.
232 e820_hole_size(unsigned long start_pfn, unsigned long end_pfn)
234 unsigned long ram = 0;
235 unsigned long start = start_pfn << PAGE_SHIFT;
236 unsigned long end = end_pfn << PAGE_SHIFT;
238 for (i = 0; i < e820.nr_map; i++) {
239 struct e820entry *ei = &e820.map[i];
240 unsigned long last, addr;
242 if (ei->type != E820_RAM ||
243 ei->addr+ei->size <= start ||
247 addr = round_up(ei->addr, PAGE_SIZE);
251 last = round_down(ei->addr + ei->size, PAGE_SIZE);
258 return ((end - start) - ram) >> PAGE_SHIFT;
262 * Mark e820 reserved areas as busy for the resource manager.
264 void __init e820_reserve_resources(void)
267 for (i = 0; i < e820.nr_map; i++) {
268 struct resource *res;
269 res = alloc_bootmem_low(sizeof(struct resource));
270 switch (e820.map[i].type) {
271 case E820_RAM: res->name = "System RAM"; break;
272 case E820_ACPI: res->name = "ACPI Tables"; break;
273 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
274 default: res->name = "reserved";
276 res->start = e820.map[i].addr;
277 res->end = res->start + e820.map[i].size - 1;
278 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
279 request_resource(&iomem_resource, res);
280 if (e820.map[i].type == E820_RAM) {
282 * We don't know which RAM region contains kernel data,
283 * so we try it repeatedly and let the resource manager
286 request_resource(res, &code_resource);
287 request_resource(res, &data_resource);
289 request_resource(res, &crashk_res);
296 * Add a memory region to the kernel e820 map.
298 void __init add_memory_region(unsigned long start, unsigned long size, int type)
303 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
307 e820.map[x].addr = start;
308 e820.map[x].size = size;
309 e820.map[x].type = type;
313 void __init e820_print_map(char *who)
317 for (i = 0; i < e820.nr_map; i++) {
318 printk(" %s: %016Lx - %016Lx ", who,
319 (unsigned long long) e820.map[i].addr,
320 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
321 switch (e820.map[i].type) {
322 case E820_RAM: printk("(usable)\n");
325 printk("(reserved)\n");
328 printk("(ACPI data)\n");
331 printk("(ACPI NVS)\n");
333 default: printk("type %u\n", e820.map[i].type);
340 * Sanitize the BIOS e820 map.
342 * Some e820 responses include overlapping entries. The following
343 * replaces the original e820 map with a new one, removing overlaps.
346 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
348 struct change_member {
349 struct e820entry *pbios; /* pointer to original bios entry */
350 unsigned long long addr; /* address for this change point */
352 static struct change_member change_point_list[2*E820MAX] __initdata;
353 static struct change_member *change_point[2*E820MAX] __initdata;
354 static struct e820entry *overlap_list[E820MAX] __initdata;
355 static struct e820entry new_bios[E820MAX] __initdata;
356 struct change_member *change_tmp;
357 unsigned long current_type, last_type;
358 unsigned long long last_addr;
359 int chgidx, still_changing;
362 int old_nr, new_nr, chg_nr;
366 Visually we're performing the following (1,2,3,4 = memory types)...
368 Sample memory map (w/overlaps):
369 ____22__________________
370 ______________________4_
371 ____1111________________
372 _44_____________________
373 11111111________________
374 ____________________33__
375 ___________44___________
376 __________33333_________
377 ______________22________
378 ___________________2222_
379 _________111111111______
380 _____________________11_
381 _________________4______
383 Sanitized equivalent (no overlap):
384 1_______________________
385 _44_____________________
386 ___1____________________
387 ____22__________________
388 ______11________________
389 _________1______________
390 __________3_____________
391 ___________44___________
392 _____________33_________
393 _______________2________
394 ________________1_______
395 _________________4______
396 ___________________2____
397 ____________________33__
398 ______________________4_
401 /* if there's only one memory region, don't bother */
407 /* bail out if we find any unreasonable addresses in bios map */
408 for (i=0; i<old_nr; i++)
409 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
412 /* create pointers for initial change-point information (for sorting) */
413 for (i=0; i < 2*old_nr; i++)
414 change_point[i] = &change_point_list[i];
416 /* record all known change-points (starting and ending addresses),
417 omitting those that are for empty memory regions */
419 for (i=0; i < old_nr; i++) {
420 if (biosmap[i].size != 0) {
421 change_point[chgidx]->addr = biosmap[i].addr;
422 change_point[chgidx++]->pbios = &biosmap[i];
423 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
424 change_point[chgidx++]->pbios = &biosmap[i];
429 /* sort change-point list by memory addresses (low -> high) */
431 while (still_changing) {
433 for (i=1; i < chg_nr; i++) {
434 /* if <current_addr> > <last_addr>, swap */
435 /* or, if current=<start_addr> & last=<end_addr>, swap */
436 if ((change_point[i]->addr < change_point[i-1]->addr) ||
437 ((change_point[i]->addr == change_point[i-1]->addr) &&
438 (change_point[i]->addr == change_point[i]->pbios->addr) &&
439 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
442 change_tmp = change_point[i];
443 change_point[i] = change_point[i-1];
444 change_point[i-1] = change_tmp;
450 /* create a new bios memory map, removing overlaps */
451 overlap_entries=0; /* number of entries in the overlap table */
452 new_bios_entry=0; /* index for creating new bios map entries */
453 last_type = 0; /* start with undefined memory type */
454 last_addr = 0; /* start with 0 as last starting address */
455 /* loop through change-points, determining affect on the new bios map */
456 for (chgidx=0; chgidx < chg_nr; chgidx++)
458 /* keep track of all overlapping bios entries */
459 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
461 /* add map entry to overlap list (> 1 entry implies an overlap) */
462 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
466 /* remove entry from list (order independent, so swap with last) */
467 for (i=0; i<overlap_entries; i++)
469 if (overlap_list[i] == change_point[chgidx]->pbios)
470 overlap_list[i] = overlap_list[overlap_entries-1];
474 /* if there are overlapping entries, decide which "type" to use */
475 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
477 for (i=0; i<overlap_entries; i++)
478 if (overlap_list[i]->type > current_type)
479 current_type = overlap_list[i]->type;
480 /* continue building up new bios map based on this information */
481 if (current_type != last_type) {
482 if (last_type != 0) {
483 new_bios[new_bios_entry].size =
484 change_point[chgidx]->addr - last_addr;
485 /* move forward only if the new size was non-zero */
486 if (new_bios[new_bios_entry].size != 0)
487 if (++new_bios_entry >= E820MAX)
488 break; /* no more space left for new bios entries */
490 if (current_type != 0) {
491 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
492 new_bios[new_bios_entry].type = current_type;
493 last_addr=change_point[chgidx]->addr;
495 last_type = current_type;
498 new_nr = new_bios_entry; /* retain count for new bios entries */
500 /* copy new bios mapping into original location */
501 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
508 * Copy the BIOS e820 map into a safe place.
510 * Sanity-check it while we're at it..
512 * If we're lucky and live on a modern system, the setup code
513 * will have given us a memory map that we can use to properly
514 * set up memory. If we aren't, we'll fake a memory map.
516 * We check to see that the memory map contains at least 2 elements
517 * before we'll use it, because the detection code in setup.S may
518 * not be perfect and most every PC known to man has two memory
519 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
520 * thinkpad 560x, for example, does not cooperate with the memory
523 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
525 /* Only one memory region (or negative)? Ignore it */
530 unsigned long start = biosmap->addr;
531 unsigned long size = biosmap->size;
532 unsigned long end = start + size;
533 unsigned long type = biosmap->type;
535 /* Overflow in 64 bits? Ignore the memory map. */
540 * Some BIOSes claim RAM in the 640k - 1M region.
541 * Not right. Fix it up.
543 * This should be removed on Hammer which is supposed to not
544 * have non e820 covered ISA mappings there, but I had some strange
545 * problems so it stays for now. -AK
547 if (type == E820_RAM) {
548 if (start < 0x100000ULL && end > 0xA0000ULL) {
549 if (start < 0xA0000ULL)
550 add_memory_region(start, 0xA0000ULL-start, type);
551 if (end <= 0x100000ULL)
558 add_memory_region(start, size, type);
559 } while (biosmap++,--nr_map);
563 void __init setup_memory_region(void)
565 char *who = "BIOS-e820";
568 * Try to copy the BIOS-supplied E820-map.
570 * Otherwise fake a memory map; one section from 0k->640k,
571 * the next section from 1mb->appropriate_mem_k
573 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
574 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
575 unsigned long mem_size;
577 /* compare results from other methods and take the greater */
578 if (ALT_MEM_K < EXT_MEM_K) {
579 mem_size = EXT_MEM_K;
582 mem_size = ALT_MEM_K;
587 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
588 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
590 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
594 void __init parse_memopt(char *p, char **from)
596 end_user_pfn = memparse(p, from);
597 end_user_pfn >>= PAGE_SHIFT;
600 void __init parse_memmapopt(char *p, char **from)
602 unsigned long long start_at, mem_size;
604 mem_size = memparse(p, from);
607 start_at = memparse(p+1, from);
608 add_memory_region(start_at, mem_size, E820_RAM);
609 } else if (*p == '#') {
610 start_at = memparse(p+1, from);
611 add_memory_region(start_at, mem_size, E820_ACPI);
612 } else if (*p == '$') {
613 start_at = memparse(p+1, from);
614 add_memory_region(start_at, mem_size, E820_RESERVED);
616 end_user_pfn = (mem_size >> PAGE_SHIFT);
621 unsigned long pci_mem_start = 0xaeedbabe;
622 EXPORT_SYMBOL(pci_mem_start);
625 * Search for the biggest gap in the low 32 bits of the e820
626 * memory space. We pass this space to PCI to assign MMIO resources
627 * for hotplug or unconfigured devices in.
628 * Hopefully the BIOS let enough space left.
630 __init void e820_setup_gap(void)
632 unsigned long gapstart, gapsize, round;
637 last = 0x100000000ull;
638 gapstart = 0x10000000;
642 unsigned long long start = e820.map[i].addr;
643 unsigned long long end = start + e820.map[i].size;
646 * Since "last" is at most 4GB, we know we'll
647 * fit in 32 bits if this condition is true
650 unsigned long gap = last - end;
663 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
664 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
665 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
669 * See how much we want to round up: start off with
670 * rounding to the next 1MB area.
673 while ((gapsize >> 4) > round)
675 /* Fun with two's complement */
676 pci_mem_start = (gapstart + round) & -round;
678 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
679 pci_mem_start, gapstart, gapsize);