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 * Getting sanitize_e820_map() in sync with i386 version by applying change:
7 * - Provisions for empty E820 memory regions (reported by certain BIOSes).
8 * Alex Achenbach <xela@slit.de>, December 2002.
9 * Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
12 #include <linux/config.h>
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/init.h>
16 #include <linux/bootmem.h>
17 #include <linux/ioport.h>
18 #include <linux/string.h>
19 #include <linux/kexec.h>
22 #include <asm/proto.h>
23 #include <asm/bootsetup.h>
28 * PFN of last memory page.
30 unsigned long 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);
77 /* XXX ramdisk image here? */
81 int __init e820_mapped(unsigned long start, unsigned long end, unsigned type)
84 for (i = 0; i < e820.nr_map; i++) {
85 struct e820entry *ei = &e820.map[i];
86 if (type && ei->type != type)
88 if (ei->addr >= end || ei->addr + ei->size <= start)
96 * Find a free area in a specific range.
98 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
101 for (i = 0; i < e820.nr_map; i++) {
102 struct e820entry *ei = &e820.map[i];
103 unsigned long addr = ei->addr, last;
104 if (ei->type != E820_RAM)
108 if (addr > ei->addr + ei->size)
110 while (bad_addr(&addr, size) && addr+size < ei->addr + ei->size)
113 if (last > ei->addr + ei->size)
123 * Free bootmem based on the e820 table for a node.
125 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
128 for (i = 0; i < e820.nr_map; i++) {
129 struct e820entry *ei = &e820.map[i];
130 unsigned long last, addr;
132 if (ei->type != E820_RAM ||
133 ei->addr+ei->size <= start ||
137 addr = round_up(ei->addr, PAGE_SIZE);
141 last = round_down(ei->addr + ei->size, PAGE_SIZE);
145 if (last > addr && last-addr >= PAGE_SIZE)
146 free_bootmem_node(pgdat, addr, last-addr);
151 * Find the highest page frame number we have available
153 unsigned long __init e820_end_of_ram(void)
156 unsigned long end_pfn = 0;
158 for (i = 0; i < e820.nr_map; i++) {
159 struct e820entry *ei = &e820.map[i];
160 unsigned long start, end;
162 start = round_up(ei->addr, PAGE_SIZE);
163 end = round_down(ei->addr + ei->size, PAGE_SIZE);
166 if (ei->type == E820_RAM) {
167 if (end > end_pfn<<PAGE_SHIFT)
168 end_pfn = end>>PAGE_SHIFT;
170 if (end > end_pfn_map<<PAGE_SHIFT)
171 end_pfn_map = end>>PAGE_SHIFT;
175 if (end_pfn > end_pfn_map)
176 end_pfn_map = end_pfn;
177 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
178 end_pfn_map = MAXMEM>>PAGE_SHIFT;
179 if (end_pfn > end_user_pfn)
180 end_pfn = end_user_pfn;
181 if (end_pfn > end_pfn_map)
182 end_pfn = end_pfn_map;
188 * Compute how much memory is missing in a range.
189 * Unlike the other functions in this file the arguments are in page numbers.
192 e820_hole_size(unsigned long start_pfn, unsigned long end_pfn)
194 unsigned long ram = 0;
195 unsigned long start = start_pfn << PAGE_SHIFT;
196 unsigned long end = end_pfn << PAGE_SHIFT;
198 for (i = 0; i < e820.nr_map; i++) {
199 struct e820entry *ei = &e820.map[i];
200 unsigned long last, addr;
202 if (ei->type != E820_RAM ||
203 ei->addr+ei->size <= start ||
207 addr = round_up(ei->addr, PAGE_SIZE);
211 last = round_down(ei->addr + ei->size, PAGE_SIZE);
218 return ((end - start) - ram) >> PAGE_SHIFT;
222 * Mark e820 reserved areas as busy for the resource manager.
224 void __init e820_reserve_resources(void)
227 for (i = 0; i < e820.nr_map; i++) {
228 struct resource *res;
229 res = alloc_bootmem_low(sizeof(struct resource));
230 switch (e820.map[i].type) {
231 case E820_RAM: res->name = "System RAM"; break;
232 case E820_ACPI: res->name = "ACPI Tables"; break;
233 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
234 default: res->name = "reserved";
236 res->start = e820.map[i].addr;
237 res->end = res->start + e820.map[i].size - 1;
238 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
239 request_resource(&iomem_resource, res);
240 if (e820.map[i].type == E820_RAM) {
242 * We don't know which RAM region contains kernel data,
243 * so we try it repeatedly and let the resource manager
246 request_resource(res, &code_resource);
247 request_resource(res, &data_resource);
249 request_resource(res, &crashk_res);
256 * Add a memory region to the kernel e820 map.
258 void __init add_memory_region(unsigned long start, unsigned long size, int type)
263 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
267 e820.map[x].addr = start;
268 e820.map[x].size = size;
269 e820.map[x].type = type;
273 void __init e820_print_map(char *who)
277 for (i = 0; i < e820.nr_map; i++) {
278 printk(" %s: %016Lx - %016Lx ", who,
279 (unsigned long long) e820.map[i].addr,
280 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
281 switch (e820.map[i].type) {
282 case E820_RAM: printk("(usable)\n");
285 printk("(reserved)\n");
288 printk("(ACPI data)\n");
291 printk("(ACPI NVS)\n");
293 default: printk("type %u\n", e820.map[i].type);
300 * Sanitize the BIOS e820 map.
302 * Some e820 responses include overlapping entries. The following
303 * replaces the original e820 map with a new one, removing overlaps.
306 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
308 struct change_member {
309 struct e820entry *pbios; /* pointer to original bios entry */
310 unsigned long long addr; /* address for this change point */
312 static struct change_member change_point_list[2*E820MAX] __initdata;
313 static struct change_member *change_point[2*E820MAX] __initdata;
314 static struct e820entry *overlap_list[E820MAX] __initdata;
315 static struct e820entry new_bios[E820MAX] __initdata;
316 struct change_member *change_tmp;
317 unsigned long current_type, last_type;
318 unsigned long long last_addr;
319 int chgidx, still_changing;
322 int old_nr, new_nr, chg_nr;
326 Visually we're performing the following (1,2,3,4 = memory types)...
328 Sample memory map (w/overlaps):
329 ____22__________________
330 ______________________4_
331 ____1111________________
332 _44_____________________
333 11111111________________
334 ____________________33__
335 ___________44___________
336 __________33333_________
337 ______________22________
338 ___________________2222_
339 _________111111111______
340 _____________________11_
341 _________________4______
343 Sanitized equivalent (no overlap):
344 1_______________________
345 _44_____________________
346 ___1____________________
347 ____22__________________
348 ______11________________
349 _________1______________
350 __________3_____________
351 ___________44___________
352 _____________33_________
353 _______________2________
354 ________________1_______
355 _________________4______
356 ___________________2____
357 ____________________33__
358 ______________________4_
361 /* if there's only one memory region, don't bother */
367 /* bail out if we find any unreasonable addresses in bios map */
368 for (i=0; i<old_nr; i++)
369 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
372 /* create pointers for initial change-point information (for sorting) */
373 for (i=0; i < 2*old_nr; i++)
374 change_point[i] = &change_point_list[i];
376 /* record all known change-points (starting and ending addresses),
377 omitting those that are for empty memory regions */
379 for (i=0; i < old_nr; i++) {
380 if (biosmap[i].size != 0) {
381 change_point[chgidx]->addr = biosmap[i].addr;
382 change_point[chgidx++]->pbios = &biosmap[i];
383 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
384 change_point[chgidx++]->pbios = &biosmap[i];
389 /* sort change-point list by memory addresses (low -> high) */
391 while (still_changing) {
393 for (i=1; i < chg_nr; i++) {
394 /* if <current_addr> > <last_addr>, swap */
395 /* or, if current=<start_addr> & last=<end_addr>, swap */
396 if ((change_point[i]->addr < change_point[i-1]->addr) ||
397 ((change_point[i]->addr == change_point[i-1]->addr) &&
398 (change_point[i]->addr == change_point[i]->pbios->addr) &&
399 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
402 change_tmp = change_point[i];
403 change_point[i] = change_point[i-1];
404 change_point[i-1] = change_tmp;
410 /* create a new bios memory map, removing overlaps */
411 overlap_entries=0; /* number of entries in the overlap table */
412 new_bios_entry=0; /* index for creating new bios map entries */
413 last_type = 0; /* start with undefined memory type */
414 last_addr = 0; /* start with 0 as last starting address */
415 /* loop through change-points, determining affect on the new bios map */
416 for (chgidx=0; chgidx < chg_nr; chgidx++)
418 /* keep track of all overlapping bios entries */
419 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
421 /* add map entry to overlap list (> 1 entry implies an overlap) */
422 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
426 /* remove entry from list (order independent, so swap with last) */
427 for (i=0; i<overlap_entries; i++)
429 if (overlap_list[i] == change_point[chgidx]->pbios)
430 overlap_list[i] = overlap_list[overlap_entries-1];
434 /* if there are overlapping entries, decide which "type" to use */
435 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
437 for (i=0; i<overlap_entries; i++)
438 if (overlap_list[i]->type > current_type)
439 current_type = overlap_list[i]->type;
440 /* continue building up new bios map based on this information */
441 if (current_type != last_type) {
442 if (last_type != 0) {
443 new_bios[new_bios_entry].size =
444 change_point[chgidx]->addr - last_addr;
445 /* move forward only if the new size was non-zero */
446 if (new_bios[new_bios_entry].size != 0)
447 if (++new_bios_entry >= E820MAX)
448 break; /* no more space left for new bios entries */
450 if (current_type != 0) {
451 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
452 new_bios[new_bios_entry].type = current_type;
453 last_addr=change_point[chgidx]->addr;
455 last_type = current_type;
458 new_nr = new_bios_entry; /* retain count for new bios entries */
460 /* copy new bios mapping into original location */
461 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
468 * Copy the BIOS e820 map into a safe place.
470 * Sanity-check it while we're at it..
472 * If we're lucky and live on a modern system, the setup code
473 * will have given us a memory map that we can use to properly
474 * set up memory. If we aren't, we'll fake a memory map.
476 * We check to see that the memory map contains at least 2 elements
477 * before we'll use it, because the detection code in setup.S may
478 * not be perfect and most every PC known to man has two memory
479 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
480 * thinkpad 560x, for example, does not cooperate with the memory
483 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
485 /* Only one memory region (or negative)? Ignore it */
490 unsigned long start = biosmap->addr;
491 unsigned long size = biosmap->size;
492 unsigned long end = start + size;
493 unsigned long type = biosmap->type;
495 /* Overflow in 64 bits? Ignore the memory map. */
500 * Some BIOSes claim RAM in the 640k - 1M region.
501 * Not right. Fix it up.
503 * This should be removed on Hammer which is supposed to not
504 * have non e820 covered ISA mappings there, but I had some strange
505 * problems so it stays for now. -AK
507 if (type == E820_RAM) {
508 if (start < 0x100000ULL && end > 0xA0000ULL) {
509 if (start < 0xA0000ULL)
510 add_memory_region(start, 0xA0000ULL-start, type);
511 if (end <= 0x100000ULL)
518 add_memory_region(start, size, type);
519 } while (biosmap++,--nr_map);
523 void __init setup_memory_region(void)
525 char *who = "BIOS-e820";
528 * Try to copy the BIOS-supplied E820-map.
530 * Otherwise fake a memory map; one section from 0k->640k,
531 * the next section from 1mb->appropriate_mem_k
533 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
534 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
535 unsigned long mem_size;
537 /* compare results from other methods and take the greater */
538 if (ALT_MEM_K < EXT_MEM_K) {
539 mem_size = EXT_MEM_K;
542 mem_size = ALT_MEM_K;
547 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
548 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
550 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
554 void __init parse_memopt(char *p, char **from)
556 end_user_pfn = memparse(p, from);
557 end_user_pfn >>= PAGE_SHIFT;
560 unsigned long pci_mem_start = 0xaeedbabe;
563 * Search for the biggest gap in the low 32 bits of the e820
564 * memory space. We pass this space to PCI to assign MMIO resources
565 * for hotplug or unconfigured devices in.
566 * Hopefully the BIOS let enough space left.
568 __init void e820_setup_gap(void)
570 unsigned long gapstart, gapsize;
575 last = 0x100000000ull;
576 gapstart = 0x10000000;
580 unsigned long long start = e820.map[i].addr;
581 unsigned long long end = start + e820.map[i].size;
584 * Since "last" is at most 4GB, we know we'll
585 * fit in 32 bits if this condition is true
588 unsigned long gap = last - end;
601 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
602 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
603 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
607 * Start allocating dynamic PCI memory a bit into the gap,
608 * aligned up to the nearest megabyte.
610 * Question: should we try to pad it up a bit (do something
611 * like " + (gapsize >> 3)" in there too?). We now have the
614 pci_mem_start = (gapstart + 0xfffff) & ~0xfffff;
616 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
617 pci_mem_start, gapstart, gapsize);