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 #include <linux/config.h>
7 #include <linux/kernel.h>
8 #include <linux/types.h>
9 #include <linux/init.h>
10 #include <linux/bootmem.h>
11 #include <linux/ioport.h>
12 #include <linux/string.h>
15 #include <asm/proto.h>
16 #include <asm/bootsetup.h>
21 * PFN of last memory page.
23 unsigned long end_pfn;
26 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
27 * The direct mapping extends to end_pfn_map, so that we can directly access
28 * apertures, ACPI and other tables without having to play with fixmaps.
30 unsigned long end_pfn_map;
33 * Last pfn which the user wants to use.
35 unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT;
37 extern struct resource code_resource, data_resource;
39 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
40 static inline int bad_addr(unsigned long *addrp, unsigned long size)
42 unsigned long addr = *addrp, last = addr + size;
44 /* various gunk below that needed for SMP startup */
50 /* direct mapping tables of the kernel */
51 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
52 *addrp = table_end << PAGE_SHIFT;
57 #ifdef CONFIG_BLK_DEV_INITRD
58 if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
59 addr < INITRD_START+INITRD_SIZE) {
60 *addrp = INITRD_START + INITRD_SIZE;
64 /* kernel code + 640k memory hole (later should not be needed, but
65 be paranoid for now) */
66 if (last >= 640*1024 && addr < __pa_symbol(&_end)) {
67 *addrp = __pa_symbol(&_end);
70 /* XXX ramdisk image here? */
74 int __init e820_mapped(unsigned long start, unsigned long end, unsigned type)
77 for (i = 0; i < e820.nr_map; i++) {
78 struct e820entry *ei = &e820.map[i];
79 if (type && ei->type != type)
81 if (ei->addr >= end || ei->addr + ei->size < start)
89 * Find a free area in a specific range.
91 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
94 for (i = 0; i < e820.nr_map; i++) {
95 struct e820entry *ei = &e820.map[i];
96 unsigned long addr = ei->addr, last;
97 if (ei->type != E820_RAM)
101 if (addr > ei->addr + ei->size)
103 while (bad_addr(&addr, size) && addr+size < ei->addr + ei->size)
106 if (last > ei->addr + ei->size)
116 * Free bootmem based on the e820 table for a node.
118 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
121 for (i = 0; i < e820.nr_map; i++) {
122 struct e820entry *ei = &e820.map[i];
123 unsigned long last, addr;
125 if (ei->type != E820_RAM ||
126 ei->addr+ei->size <= start ||
130 addr = round_up(ei->addr, PAGE_SIZE);
134 last = round_down(ei->addr + ei->size, PAGE_SIZE);
138 if (last > addr && last-addr >= PAGE_SIZE)
139 free_bootmem_node(pgdat, addr, last-addr);
144 * Find the highest page frame number we have available
146 unsigned long __init e820_end_of_ram(void)
149 unsigned long end_pfn = 0;
151 for (i = 0; i < e820.nr_map; i++) {
152 struct e820entry *ei = &e820.map[i];
153 unsigned long start, end;
155 start = round_up(ei->addr, PAGE_SIZE);
156 end = round_down(ei->addr + ei->size, PAGE_SIZE);
159 if (ei->type == E820_RAM) {
160 if (end > end_pfn<<PAGE_SHIFT)
161 end_pfn = end>>PAGE_SHIFT;
163 if (end > end_pfn_map<<PAGE_SHIFT)
164 end_pfn_map = end>>PAGE_SHIFT;
168 if (end_pfn > end_pfn_map)
169 end_pfn_map = end_pfn;
170 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
171 end_pfn_map = MAXMEM>>PAGE_SHIFT;
172 if (end_pfn > end_user_pfn)
173 end_pfn = end_user_pfn;
174 if (end_pfn > end_pfn_map)
175 end_pfn = end_pfn_map;
181 * Mark e820 reserved areas as busy for the resource manager.
183 void __init e820_reserve_resources(void)
186 for (i = 0; i < e820.nr_map; i++) {
187 struct resource *res;
188 res = alloc_bootmem_low(sizeof(struct resource));
189 switch (e820.map[i].type) {
190 case E820_RAM: res->name = "System RAM"; break;
191 case E820_ACPI: res->name = "ACPI Tables"; break;
192 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
193 default: res->name = "reserved";
195 res->start = e820.map[i].addr;
196 res->end = res->start + e820.map[i].size - 1;
197 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
198 request_resource(&iomem_resource, res);
199 if (e820.map[i].type == E820_RAM) {
201 * We don't know which RAM region contains kernel data,
202 * so we try it repeatedly and let the resource manager
205 request_resource(res, &code_resource);
206 request_resource(res, &data_resource);
212 * Add a memory region to the kernel e820 map.
214 void __init add_memory_region(unsigned long start, unsigned long size, int type)
219 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
223 e820.map[x].addr = start;
224 e820.map[x].size = size;
225 e820.map[x].type = type;
229 void __init e820_print_map(char *who)
233 for (i = 0; i < e820.nr_map; i++) {
234 printk(" %s: %016Lx - %016Lx ", who,
235 (unsigned long long) e820.map[i].addr,
236 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
237 switch (e820.map[i].type) {
238 case E820_RAM: printk("(usable)\n");
241 printk("(reserved)\n");
244 printk("(ACPI data)\n");
247 printk("(ACPI NVS)\n");
249 default: printk("type %u\n", e820.map[i].type);
256 * Sanitize the BIOS e820 map.
258 * Some e820 responses include overlapping entries. The following
259 * replaces the original e820 map with a new one, removing overlaps.
262 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
264 struct change_member {
265 struct e820entry *pbios; /* pointer to original bios entry */
266 unsigned long long addr; /* address for this change point */
268 static struct change_member change_point_list[2*E820MAX] __initdata;
269 static struct change_member *change_point[2*E820MAX] __initdata;
270 static struct e820entry *overlap_list[E820MAX] __initdata;
271 static struct e820entry new_bios[E820MAX] __initdata;
272 struct change_member *change_tmp;
273 unsigned long current_type, last_type;
274 unsigned long long last_addr;
275 int chgidx, still_changing;
282 Visually we're performing the following (1,2,3,4 = memory types)...
284 Sample memory map (w/overlaps):
285 ____22__________________
286 ______________________4_
287 ____1111________________
288 _44_____________________
289 11111111________________
290 ____________________33__
291 ___________44___________
292 __________33333_________
293 ______________22________
294 ___________________2222_
295 _________111111111______
296 _____________________11_
297 _________________4______
299 Sanitized equivalent (no overlap):
300 1_______________________
301 _44_____________________
302 ___1____________________
303 ____22__________________
304 ______11________________
305 _________1______________
306 __________3_____________
307 ___________44___________
308 _____________33_________
309 _______________2________
310 ________________1_______
311 _________________4______
312 ___________________2____
313 ____________________33__
314 ______________________4_
317 /* if there's only one memory region, don't bother */
323 /* bail out if we find any unreasonable addresses in bios map */
324 for (i=0; i<old_nr; i++)
325 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
328 /* create pointers for initial change-point information (for sorting) */
329 for (i=0; i < 2*old_nr; i++)
330 change_point[i] = &change_point_list[i];
332 /* record all known change-points (starting and ending addresses) */
334 for (i=0; i < old_nr; i++) {
335 change_point[chgidx]->addr = biosmap[i].addr;
336 change_point[chgidx++]->pbios = &biosmap[i];
337 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
338 change_point[chgidx++]->pbios = &biosmap[i];
341 /* sort change-point list by memory addresses (low -> high) */
343 while (still_changing) {
345 for (i=1; i < 2*old_nr; i++) {
346 /* if <current_addr> > <last_addr>, swap */
347 /* or, if current=<start_addr> & last=<end_addr>, swap */
348 if ((change_point[i]->addr < change_point[i-1]->addr) ||
349 ((change_point[i]->addr == change_point[i-1]->addr) &&
350 (change_point[i]->addr == change_point[i]->pbios->addr) &&
351 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
354 change_tmp = change_point[i];
355 change_point[i] = change_point[i-1];
356 change_point[i-1] = change_tmp;
362 /* create a new bios memory map, removing overlaps */
363 overlap_entries=0; /* number of entries in the overlap table */
364 new_bios_entry=0; /* index for creating new bios map entries */
365 last_type = 0; /* start with undefined memory type */
366 last_addr = 0; /* start with 0 as last starting address */
367 /* loop through change-points, determining affect on the new bios map */
368 for (chgidx=0; chgidx < 2*old_nr; chgidx++)
370 /* keep track of all overlapping bios entries */
371 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
373 /* add map entry to overlap list (> 1 entry implies an overlap) */
374 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
378 /* remove entry from list (order independent, so swap with last) */
379 for (i=0; i<overlap_entries; i++)
381 if (overlap_list[i] == change_point[chgidx]->pbios)
382 overlap_list[i] = overlap_list[overlap_entries-1];
386 /* if there are overlapping entries, decide which "type" to use */
387 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
389 for (i=0; i<overlap_entries; i++)
390 if (overlap_list[i]->type > current_type)
391 current_type = overlap_list[i]->type;
392 /* continue building up new bios map based on this information */
393 if (current_type != last_type) {
394 if (last_type != 0) {
395 new_bios[new_bios_entry].size =
396 change_point[chgidx]->addr - last_addr;
397 /* move forward only if the new size was non-zero */
398 if (new_bios[new_bios_entry].size != 0)
399 if (++new_bios_entry >= E820MAX)
400 break; /* no more space left for new bios entries */
402 if (current_type != 0) {
403 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
404 new_bios[new_bios_entry].type = current_type;
405 last_addr=change_point[chgidx]->addr;
407 last_type = current_type;
410 new_nr = new_bios_entry; /* retain count for new bios entries */
412 /* copy new bios mapping into original location */
413 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
420 * Copy the BIOS e820 map into a safe place.
422 * Sanity-check it while we're at it..
424 * If we're lucky and live on a modern system, the setup code
425 * will have given us a memory map that we can use to properly
426 * set up memory. If we aren't, we'll fake a memory map.
428 * We check to see that the memory map contains at least 2 elements
429 * before we'll use it, because the detection code in setup.S may
430 * not be perfect and most every PC known to man has two memory
431 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
432 * thinkpad 560x, for example, does not cooperate with the memory
435 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
437 /* Only one memory region (or negative)? Ignore it */
442 unsigned long start = biosmap->addr;
443 unsigned long size = biosmap->size;
444 unsigned long end = start + size;
445 unsigned long type = biosmap->type;
447 /* Overflow in 64 bits? Ignore the memory map. */
452 * Some BIOSes claim RAM in the 640k - 1M region.
453 * Not right. Fix it up.
455 * This should be removed on Hammer which is supposed to not
456 * have non e820 covered ISA mappings there, but I had some strange
457 * problems so it stays for now. -AK
459 if (type == E820_RAM) {
460 if (start < 0x100000ULL && end > 0xA0000ULL) {
461 if (start < 0xA0000ULL)
462 add_memory_region(start, 0xA0000ULL-start, type);
463 if (end <= 0x100000ULL)
470 add_memory_region(start, size, type);
471 } while (biosmap++,--nr_map);
475 void __init setup_memory_region(void)
477 char *who = "BIOS-e820";
480 * Try to copy the BIOS-supplied E820-map.
482 * Otherwise fake a memory map; one section from 0k->640k,
483 * the next section from 1mb->appropriate_mem_k
485 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
486 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
487 unsigned long mem_size;
489 /* compare results from other methods and take the greater */
490 if (ALT_MEM_K < EXT_MEM_K) {
491 mem_size = EXT_MEM_K;
494 mem_size = ALT_MEM_K;
499 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
500 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
502 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
506 void __init parse_memopt(char *p, char **from)
509 * mem=XXX[kKmM] limits kernel memory to XXX+1MB
511 * It would be more logical to count from 0 instead of from
512 * HIGH_MEMORY, but we keep that for now for i386 compatibility.
514 * No support for custom mapping like i386. The reason is
515 * that we need to read the e820 map anyways to handle the
516 * ACPI mappings in the direct map. Also on x86-64 there
517 * should be always a good e820 map. This is only an upper
518 * limit, you cannot force usage of memory not in e820.
522 end_user_pfn = memparse(p, from) + HIGH_MEMORY;
523 end_user_pfn >>= PAGE_SHIFT;