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 < __pa_symbol(&_end)) {
75 *addrp = __pa_symbol(&_end);
79 if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
80 *addrp = ebda_addr + ebda_size;
84 /* XXX ramdisk image here? */
89 * This function checks if any part of the range <start,end> is mapped
93 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
96 for (i = 0; i < e820.nr_map; i++) {
97 struct e820entry *ei = &e820.map[i];
98 if (type && ei->type != type)
100 if (ei->addr >= end || ei->addr + ei->size <= start)
108 * Find a free area in a specific range.
110 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
113 for (i = 0; i < e820.nr_map; i++) {
114 struct e820entry *ei = &e820.map[i];
115 unsigned long addr = ei->addr, last;
116 if (ei->type != E820_RAM)
120 if (addr > ei->addr + ei->size)
122 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
125 if (last > ei->addr + ei->size)
135 * Free bootmem based on the e820 table for a node.
137 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
140 for (i = 0; i < e820.nr_map; i++) {
141 struct e820entry *ei = &e820.map[i];
142 unsigned long last, addr;
144 if (ei->type != E820_RAM ||
145 ei->addr+ei->size <= start ||
149 addr = round_up(ei->addr, PAGE_SIZE);
153 last = round_down(ei->addr + ei->size, PAGE_SIZE);
157 if (last > addr && last-addr >= PAGE_SIZE)
158 free_bootmem_node(pgdat, addr, last-addr);
163 * Find the highest page frame number we have available
165 unsigned long __init e820_end_of_ram(void)
168 unsigned long end_pfn = 0;
170 for (i = 0; i < e820.nr_map; i++) {
171 struct e820entry *ei = &e820.map[i];
172 unsigned long start, end;
174 start = round_up(ei->addr, PAGE_SIZE);
175 end = round_down(ei->addr + ei->size, PAGE_SIZE);
178 if (ei->type == E820_RAM) {
179 if (end > end_pfn<<PAGE_SHIFT)
180 end_pfn = end>>PAGE_SHIFT;
182 if (end > end_pfn_map<<PAGE_SHIFT)
183 end_pfn_map = end>>PAGE_SHIFT;
187 if (end_pfn > end_pfn_map)
188 end_pfn_map = end_pfn;
189 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
190 end_pfn_map = MAXMEM>>PAGE_SHIFT;
191 if (end_pfn > end_user_pfn)
192 end_pfn = end_user_pfn;
193 if (end_pfn > end_pfn_map)
194 end_pfn = end_pfn_map;
200 * Compute how much memory is missing in a range.
201 * Unlike the other functions in this file the arguments are in page numbers.
204 e820_hole_size(unsigned long start_pfn, unsigned long end_pfn)
206 unsigned long ram = 0;
207 unsigned long start = start_pfn << PAGE_SHIFT;
208 unsigned long end = end_pfn << PAGE_SHIFT;
210 for (i = 0; i < e820.nr_map; i++) {
211 struct e820entry *ei = &e820.map[i];
212 unsigned long last, addr;
214 if (ei->type != E820_RAM ||
215 ei->addr+ei->size <= start ||
219 addr = round_up(ei->addr, PAGE_SIZE);
223 last = round_down(ei->addr + ei->size, PAGE_SIZE);
230 return ((end - start) - ram) >> PAGE_SHIFT;
234 * Mark e820 reserved areas as busy for the resource manager.
236 void __init e820_reserve_resources(void)
239 for (i = 0; i < e820.nr_map; i++) {
240 struct resource *res;
241 res = alloc_bootmem_low(sizeof(struct resource));
242 switch (e820.map[i].type) {
243 case E820_RAM: res->name = "System RAM"; break;
244 case E820_ACPI: res->name = "ACPI Tables"; break;
245 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
246 default: res->name = "reserved";
248 res->start = e820.map[i].addr;
249 res->end = res->start + e820.map[i].size - 1;
250 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
251 request_resource(&iomem_resource, res);
252 if (e820.map[i].type == E820_RAM) {
254 * We don't know which RAM region contains kernel data,
255 * so we try it repeatedly and let the resource manager
258 request_resource(res, &code_resource);
259 request_resource(res, &data_resource);
261 request_resource(res, &crashk_res);
268 * Add a memory region to the kernel e820 map.
270 void __init add_memory_region(unsigned long start, unsigned long size, int type)
275 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
279 e820.map[x].addr = start;
280 e820.map[x].size = size;
281 e820.map[x].type = type;
285 void __init e820_print_map(char *who)
289 for (i = 0; i < e820.nr_map; i++) {
290 printk(" %s: %016Lx - %016Lx ", who,
291 (unsigned long long) e820.map[i].addr,
292 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
293 switch (e820.map[i].type) {
294 case E820_RAM: printk("(usable)\n");
297 printk("(reserved)\n");
300 printk("(ACPI data)\n");
303 printk("(ACPI NVS)\n");
305 default: printk("type %u\n", e820.map[i].type);
312 * Sanitize the BIOS e820 map.
314 * Some e820 responses include overlapping entries. The following
315 * replaces the original e820 map with a new one, removing overlaps.
318 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
320 struct change_member {
321 struct e820entry *pbios; /* pointer to original bios entry */
322 unsigned long long addr; /* address for this change point */
324 static struct change_member change_point_list[2*E820MAX] __initdata;
325 static struct change_member *change_point[2*E820MAX] __initdata;
326 static struct e820entry *overlap_list[E820MAX] __initdata;
327 static struct e820entry new_bios[E820MAX] __initdata;
328 struct change_member *change_tmp;
329 unsigned long current_type, last_type;
330 unsigned long long last_addr;
331 int chgidx, still_changing;
334 int old_nr, new_nr, chg_nr;
338 Visually we're performing the following (1,2,3,4 = memory types)...
340 Sample memory map (w/overlaps):
341 ____22__________________
342 ______________________4_
343 ____1111________________
344 _44_____________________
345 11111111________________
346 ____________________33__
347 ___________44___________
348 __________33333_________
349 ______________22________
350 ___________________2222_
351 _________111111111______
352 _____________________11_
353 _________________4______
355 Sanitized equivalent (no overlap):
356 1_______________________
357 _44_____________________
358 ___1____________________
359 ____22__________________
360 ______11________________
361 _________1______________
362 __________3_____________
363 ___________44___________
364 _____________33_________
365 _______________2________
366 ________________1_______
367 _________________4______
368 ___________________2____
369 ____________________33__
370 ______________________4_
373 /* if there's only one memory region, don't bother */
379 /* bail out if we find any unreasonable addresses in bios map */
380 for (i=0; i<old_nr; i++)
381 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
384 /* create pointers for initial change-point information (for sorting) */
385 for (i=0; i < 2*old_nr; i++)
386 change_point[i] = &change_point_list[i];
388 /* record all known change-points (starting and ending addresses),
389 omitting those that are for empty memory regions */
391 for (i=0; i < old_nr; i++) {
392 if (biosmap[i].size != 0) {
393 change_point[chgidx]->addr = biosmap[i].addr;
394 change_point[chgidx++]->pbios = &biosmap[i];
395 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
396 change_point[chgidx++]->pbios = &biosmap[i];
401 /* sort change-point list by memory addresses (low -> high) */
403 while (still_changing) {
405 for (i=1; i < chg_nr; i++) {
406 /* if <current_addr> > <last_addr>, swap */
407 /* or, if current=<start_addr> & last=<end_addr>, swap */
408 if ((change_point[i]->addr < change_point[i-1]->addr) ||
409 ((change_point[i]->addr == change_point[i-1]->addr) &&
410 (change_point[i]->addr == change_point[i]->pbios->addr) &&
411 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
414 change_tmp = change_point[i];
415 change_point[i] = change_point[i-1];
416 change_point[i-1] = change_tmp;
422 /* create a new bios memory map, removing overlaps */
423 overlap_entries=0; /* number of entries in the overlap table */
424 new_bios_entry=0; /* index for creating new bios map entries */
425 last_type = 0; /* start with undefined memory type */
426 last_addr = 0; /* start with 0 as last starting address */
427 /* loop through change-points, determining affect on the new bios map */
428 for (chgidx=0; chgidx < chg_nr; chgidx++)
430 /* keep track of all overlapping bios entries */
431 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
433 /* add map entry to overlap list (> 1 entry implies an overlap) */
434 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
438 /* remove entry from list (order independent, so swap with last) */
439 for (i=0; i<overlap_entries; i++)
441 if (overlap_list[i] == change_point[chgidx]->pbios)
442 overlap_list[i] = overlap_list[overlap_entries-1];
446 /* if there are overlapping entries, decide which "type" to use */
447 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
449 for (i=0; i<overlap_entries; i++)
450 if (overlap_list[i]->type > current_type)
451 current_type = overlap_list[i]->type;
452 /* continue building up new bios map based on this information */
453 if (current_type != last_type) {
454 if (last_type != 0) {
455 new_bios[new_bios_entry].size =
456 change_point[chgidx]->addr - last_addr;
457 /* move forward only if the new size was non-zero */
458 if (new_bios[new_bios_entry].size != 0)
459 if (++new_bios_entry >= E820MAX)
460 break; /* no more space left for new bios entries */
462 if (current_type != 0) {
463 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
464 new_bios[new_bios_entry].type = current_type;
465 last_addr=change_point[chgidx]->addr;
467 last_type = current_type;
470 new_nr = new_bios_entry; /* retain count for new bios entries */
472 /* copy new bios mapping into original location */
473 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
480 * Copy the BIOS e820 map into a safe place.
482 * Sanity-check it while we're at it..
484 * If we're lucky and live on a modern system, the setup code
485 * will have given us a memory map that we can use to properly
486 * set up memory. If we aren't, we'll fake a memory map.
488 * We check to see that the memory map contains at least 2 elements
489 * before we'll use it, because the detection code in setup.S may
490 * not be perfect and most every PC known to man has two memory
491 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
492 * thinkpad 560x, for example, does not cooperate with the memory
495 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
497 /* Only one memory region (or negative)? Ignore it */
502 unsigned long start = biosmap->addr;
503 unsigned long size = biosmap->size;
504 unsigned long end = start + size;
505 unsigned long type = biosmap->type;
507 /* Overflow in 64 bits? Ignore the memory map. */
512 * Some BIOSes claim RAM in the 640k - 1M region.
513 * Not right. Fix it up.
515 * This should be removed on Hammer which is supposed to not
516 * have non e820 covered ISA mappings there, but I had some strange
517 * problems so it stays for now. -AK
519 if (type == E820_RAM) {
520 if (start < 0x100000ULL && end > 0xA0000ULL) {
521 if (start < 0xA0000ULL)
522 add_memory_region(start, 0xA0000ULL-start, type);
523 if (end <= 0x100000ULL)
530 add_memory_region(start, size, type);
531 } while (biosmap++,--nr_map);
535 void __init setup_memory_region(void)
537 char *who = "BIOS-e820";
540 * Try to copy the BIOS-supplied E820-map.
542 * Otherwise fake a memory map; one section from 0k->640k,
543 * the next section from 1mb->appropriate_mem_k
545 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
546 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
547 unsigned long mem_size;
549 /* compare results from other methods and take the greater */
550 if (ALT_MEM_K < EXT_MEM_K) {
551 mem_size = EXT_MEM_K;
554 mem_size = ALT_MEM_K;
559 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
560 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
562 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
566 void __init parse_memopt(char *p, char **from)
568 end_user_pfn = memparse(p, from);
569 end_user_pfn >>= PAGE_SHIFT;
572 void __init parse_memmapopt(char *p, char **from)
574 unsigned long long start_at, mem_size;
576 mem_size = memparse(p, from);
579 start_at = memparse(p+1, from);
580 add_memory_region(start_at, mem_size, E820_RAM);
581 } else if (*p == '#') {
582 start_at = memparse(p+1, from);
583 add_memory_region(start_at, mem_size, E820_ACPI);
584 } else if (*p == '$') {
585 start_at = memparse(p+1, from);
586 add_memory_region(start_at, mem_size, E820_RESERVED);
588 end_user_pfn = (mem_size >> PAGE_SHIFT);
593 unsigned long pci_mem_start = 0xaeedbabe;
594 EXPORT_SYMBOL(pci_mem_start);
597 * Search for the biggest gap in the low 32 bits of the e820
598 * memory space. We pass this space to PCI to assign MMIO resources
599 * for hotplug or unconfigured devices in.
600 * Hopefully the BIOS let enough space left.
602 __init void e820_setup_gap(void)
604 unsigned long gapstart, gapsize, round;
609 last = 0x100000000ull;
610 gapstart = 0x10000000;
614 unsigned long long start = e820.map[i].addr;
615 unsigned long long end = start + e820.map[i].size;
618 * Since "last" is at most 4GB, we know we'll
619 * fit in 32 bits if this condition is true
622 unsigned long gap = last - end;
635 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
636 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
637 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
641 * See how much we want to round up: start off with
642 * rounding to the next 1MB area.
645 while ((gapsize >> 4) > round)
647 /* Fun with two's complement */
648 pci_mem_start = (gapstart + round) & -round;
650 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
651 pci_mem_start, gapstart, gapsize);