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>
20 #include <asm/pgtable.h>
23 #include <asm/proto.h>
24 #include <asm/bootsetup.h>
25 #include <asm/sections.h>
28 * PFN of last memory page.
30 unsigned long end_pfn;
31 EXPORT_SYMBOL(end_pfn);
34 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
35 * The direct mapping extends to end_pfn_map, so that we can directly access
36 * apertures, ACPI and other tables without having to play with fixmaps.
38 unsigned long end_pfn_map;
41 * Last pfn which the user wants to use.
43 unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT;
45 extern struct resource code_resource, data_resource;
47 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
48 static inline int bad_addr(unsigned long *addrp, unsigned long size)
50 unsigned long addr = *addrp, last = addr + size;
52 /* various gunk below that needed for SMP startup */
58 /* direct mapping tables of the kernel */
59 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
60 *addrp = table_end << PAGE_SHIFT;
65 #ifdef CONFIG_BLK_DEV_INITRD
66 if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
67 addr < INITRD_START+INITRD_SIZE) {
68 *addrp = INITRD_START + INITRD_SIZE;
72 /* kernel code + 640k memory hole (later should not be needed, but
73 be paranoid for now) */
74 if (last >= 640*1024 && addr < 1024*1024) {
78 if (last >= __pa_symbol(&_text) && last < __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(pg_data_t *pgdat, 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_node(pgdat, 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_low(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 for (i = 0; i < e820.nr_map; i++) {
323 printk(" %s: %016Lx - %016Lx ", who,
324 (unsigned long long) e820.map[i].addr,
325 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
326 switch (e820.map[i].type) {
327 case E820_RAM: printk("(usable)\n");
330 printk("(reserved)\n");
333 printk("(ACPI data)\n");
336 printk("(ACPI NVS)\n");
338 default: printk("type %u\n", e820.map[i].type);
345 * Sanitize the BIOS e820 map.
347 * Some e820 responses include overlapping entries. The following
348 * replaces the original e820 map with a new one, removing overlaps.
351 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
353 struct change_member {
354 struct e820entry *pbios; /* pointer to original bios entry */
355 unsigned long long addr; /* address for this change point */
357 static struct change_member change_point_list[2*E820MAX] __initdata;
358 static struct change_member *change_point[2*E820MAX] __initdata;
359 static struct e820entry *overlap_list[E820MAX] __initdata;
360 static struct e820entry new_bios[E820MAX] __initdata;
361 struct change_member *change_tmp;
362 unsigned long current_type, last_type;
363 unsigned long long last_addr;
364 int chgidx, still_changing;
367 int old_nr, new_nr, chg_nr;
371 Visually we're performing the following (1,2,3,4 = memory types)...
373 Sample memory map (w/overlaps):
374 ____22__________________
375 ______________________4_
376 ____1111________________
377 _44_____________________
378 11111111________________
379 ____________________33__
380 ___________44___________
381 __________33333_________
382 ______________22________
383 ___________________2222_
384 _________111111111______
385 _____________________11_
386 _________________4______
388 Sanitized equivalent (no overlap):
389 1_______________________
390 _44_____________________
391 ___1____________________
392 ____22__________________
393 ______11________________
394 _________1______________
395 __________3_____________
396 ___________44___________
397 _____________33_________
398 _______________2________
399 ________________1_______
400 _________________4______
401 ___________________2____
402 ____________________33__
403 ______________________4_
406 /* if there's only one memory region, don't bother */
412 /* bail out if we find any unreasonable addresses in bios map */
413 for (i=0; i<old_nr; i++)
414 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
417 /* create pointers for initial change-point information (for sorting) */
418 for (i=0; i < 2*old_nr; i++)
419 change_point[i] = &change_point_list[i];
421 /* record all known change-points (starting and ending addresses),
422 omitting those that are for empty memory regions */
424 for (i=0; i < old_nr; i++) {
425 if (biosmap[i].size != 0) {
426 change_point[chgidx]->addr = biosmap[i].addr;
427 change_point[chgidx++]->pbios = &biosmap[i];
428 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
429 change_point[chgidx++]->pbios = &biosmap[i];
434 /* sort change-point list by memory addresses (low -> high) */
436 while (still_changing) {
438 for (i=1; i < chg_nr; i++) {
439 /* if <current_addr> > <last_addr>, swap */
440 /* or, if current=<start_addr> & last=<end_addr>, swap */
441 if ((change_point[i]->addr < change_point[i-1]->addr) ||
442 ((change_point[i]->addr == change_point[i-1]->addr) &&
443 (change_point[i]->addr == change_point[i]->pbios->addr) &&
444 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
447 change_tmp = change_point[i];
448 change_point[i] = change_point[i-1];
449 change_point[i-1] = change_tmp;
455 /* create a new bios memory map, removing overlaps */
456 overlap_entries=0; /* number of entries in the overlap table */
457 new_bios_entry=0; /* index for creating new bios map entries */
458 last_type = 0; /* start with undefined memory type */
459 last_addr = 0; /* start with 0 as last starting address */
460 /* loop through change-points, determining affect on the new bios map */
461 for (chgidx=0; chgidx < chg_nr; chgidx++)
463 /* keep track of all overlapping bios entries */
464 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
466 /* add map entry to overlap list (> 1 entry implies an overlap) */
467 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
471 /* remove entry from list (order independent, so swap with last) */
472 for (i=0; i<overlap_entries; i++)
474 if (overlap_list[i] == change_point[chgidx]->pbios)
475 overlap_list[i] = overlap_list[overlap_entries-1];
479 /* if there are overlapping entries, decide which "type" to use */
480 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
482 for (i=0; i<overlap_entries; i++)
483 if (overlap_list[i]->type > current_type)
484 current_type = overlap_list[i]->type;
485 /* continue building up new bios map based on this information */
486 if (current_type != last_type) {
487 if (last_type != 0) {
488 new_bios[new_bios_entry].size =
489 change_point[chgidx]->addr - last_addr;
490 /* move forward only if the new size was non-zero */
491 if (new_bios[new_bios_entry].size != 0)
492 if (++new_bios_entry >= E820MAX)
493 break; /* no more space left for new bios entries */
495 if (current_type != 0) {
496 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
497 new_bios[new_bios_entry].type = current_type;
498 last_addr=change_point[chgidx]->addr;
500 last_type = current_type;
503 new_nr = new_bios_entry; /* retain count for new bios entries */
505 /* copy new bios mapping into original location */
506 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
513 * Copy the BIOS e820 map into a safe place.
515 * Sanity-check it while we're at it..
517 * If we're lucky and live on a modern system, the setup code
518 * will have given us a memory map that we can use to properly
519 * set up memory. If we aren't, we'll fake a memory map.
521 * We check to see that the memory map contains at least 2 elements
522 * before we'll use it, because the detection code in setup.S may
523 * not be perfect and most every PC known to man has two memory
524 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
525 * thinkpad 560x, for example, does not cooperate with the memory
528 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
530 /* Only one memory region (or negative)? Ignore it */
535 unsigned long start = biosmap->addr;
536 unsigned long size = biosmap->size;
537 unsigned long end = start + size;
538 unsigned long type = biosmap->type;
540 /* Overflow in 64 bits? Ignore the memory map. */
545 * Some BIOSes claim RAM in the 640k - 1M region.
546 * Not right. Fix it up.
548 * This should be removed on Hammer which is supposed to not
549 * have non e820 covered ISA mappings there, but I had some strange
550 * problems so it stays for now. -AK
552 if (type == E820_RAM) {
553 if (start < 0x100000ULL && end > 0xA0000ULL) {
554 if (start < 0xA0000ULL)
555 add_memory_region(start, 0xA0000ULL-start, type);
556 if (end <= 0x100000ULL)
563 add_memory_region(start, size, type);
564 } while (biosmap++,--nr_map);
568 void __init setup_memory_region(void)
570 char *who = "BIOS-e820";
573 * Try to copy the BIOS-supplied E820-map.
575 * Otherwise fake a memory map; one section from 0k->640k,
576 * the next section from 1mb->appropriate_mem_k
578 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
579 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
580 unsigned long mem_size;
582 /* compare results from other methods and take the greater */
583 if (ALT_MEM_K < EXT_MEM_K) {
584 mem_size = EXT_MEM_K;
587 mem_size = ALT_MEM_K;
592 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
593 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
595 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
599 void __init parse_memopt(char *p, char **from)
601 end_user_pfn = memparse(p, from);
602 end_user_pfn >>= PAGE_SHIFT;
605 void __init parse_memmapopt(char *p, char **from)
607 unsigned long long start_at, mem_size;
609 mem_size = memparse(p, from);
612 start_at = memparse(p+1, from);
613 add_memory_region(start_at, mem_size, E820_RAM);
614 } else if (*p == '#') {
615 start_at = memparse(p+1, from);
616 add_memory_region(start_at, mem_size, E820_ACPI);
617 } else if (*p == '$') {
618 start_at = memparse(p+1, from);
619 add_memory_region(start_at, mem_size, E820_RESERVED);
621 end_user_pfn = (mem_size >> PAGE_SHIFT);
626 unsigned long pci_mem_start = 0xaeedbabe;
627 EXPORT_SYMBOL(pci_mem_start);
630 * Search for the biggest gap in the low 32 bits of the e820
631 * memory space. We pass this space to PCI to assign MMIO resources
632 * for hotplug or unconfigured devices in.
633 * Hopefully the BIOS let enough space left.
635 __init void e820_setup_gap(void)
637 unsigned long gapstart, gapsize, round;
642 last = 0x100000000ull;
643 gapstart = 0x10000000;
647 unsigned long long start = e820.map[i].addr;
648 unsigned long long end = start + e820.map[i].size;
651 * Since "last" is at most 4GB, we know we'll
652 * fit in 32 bits if this condition is true
655 unsigned long gap = last - end;
668 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
669 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
670 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
674 * See how much we want to round up: start off with
675 * rounding to the next 1MB area.
678 while ((gapsize >> 4) > round)
680 /* Fun with two's complement */
681 pci_mem_start = (gapstart + round) & -round;
683 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
684 pci_mem_start, gapstart, gapsize);