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>
20 #include <linux/module.h>
24 #include <asm/proto.h>
25 #include <asm/bootsetup.h>
26 #include <asm/sections.h>
27 #include <xen/interface/memory.h>
30 * PFN of last memory page.
32 unsigned long end_pfn;
33 EXPORT_SYMBOL(end_pfn);
36 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
37 * The direct mapping extends to end_pfn_map, so that we can directly access
38 * apertures, ACPI and other tables without having to play with fixmaps.
40 unsigned long end_pfn_map;
43 * Last pfn which the user wants to use.
45 unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT;
47 extern struct resource code_resource, data_resource;
49 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
50 static inline int bad_addr(unsigned long *addrp, unsigned long size)
52 unsigned long addr = *addrp, last = addr + size;
55 /* various gunk below that needed for SMP startup */
61 /* direct mapping tables of the kernel */
62 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
63 *addrp = table_end << PAGE_SHIFT;
68 #ifdef CONFIG_BLK_DEV_INITRD
69 if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
70 addr < INITRD_START+INITRD_SIZE) {
71 *addrp = INITRD_START + INITRD_SIZE;
75 /* kernel code + 640k memory hole (later should not be needed, but
76 be paranoid for now) */
77 if (last >= 640*1024 && addr < __pa_symbol(&_end)) {
78 *addrp = __pa_symbol(&_end);
82 if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
83 *addrp = ebda_addr + ebda_size;
87 /* XXX ramdisk image here? */
89 if (last < (table_end<<PAGE_SHIFT)) {
90 *addrp = table_end << PAGE_SHIFT;
99 * This function checks if any part of the range <start,end> is mapped
103 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
106 for (i = 0; i < e820.nr_map; i++) {
107 struct e820entry *ei = &e820.map[i];
108 if (type && ei->type != type)
110 if (ei->addr >= end || ei->addr + ei->size <= start)
119 * This function checks if the entire range <start,end> is mapped with type.
121 * Note: this function only works correct if the e820 table is sorted and
122 * not-overlapping, which is the case
124 int __init e820_all_mapped(unsigned long start, unsigned long end, unsigned type)
127 for (i = 0; i < e820.nr_map; i++) {
128 struct e820entry *ei = &e820.map[i];
129 if (type && ei->type != type)
131 /* is the region (part) in overlap with the current region ?*/
132 if (ei->addr >= end || ei->addr + ei->size <= start)
135 /* if the region is at the beginning of <start,end> we move
136 * start to the end of the region since it's ok until there
138 if (ei->addr <= start)
139 start = ei->addr + ei->size;
140 /* if start is now at or beyond end, we're done, full coverage */
142 return 1; /* we're done */
148 * Find a free area in a specific range.
150 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
153 for (i = 0; i < e820.nr_map; i++) {
154 struct e820entry *ei = &e820.map[i];
155 unsigned long addr = ei->addr, last;
156 if (ei->type != E820_RAM)
160 if (addr > ei->addr + ei->size)
162 while (bad_addr(&addr, size) && addr+size < ei->addr + ei->size)
165 if (last > ei->addr + ei->size)
175 * Free bootmem based on the e820 table for a node.
177 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
180 for (i = 0; i < e820.nr_map; i++) {
181 struct e820entry *ei = &e820.map[i];
182 unsigned long last, addr;
184 if (ei->type != E820_RAM ||
185 ei->addr+ei->size <= start ||
189 addr = round_up(ei->addr, PAGE_SIZE);
193 last = round_down(ei->addr + ei->size, PAGE_SIZE);
197 if (last > addr && last-addr >= PAGE_SIZE)
198 free_bootmem_node(pgdat, addr, last-addr);
203 * Find the highest page frame number we have available
205 unsigned long __init e820_end_of_ram(void)
208 unsigned long end_pfn = 0;
210 for (i = 0; i < e820.nr_map; i++) {
211 struct e820entry *ei = &e820.map[i];
212 unsigned long start, end;
214 start = round_up(ei->addr, PAGE_SIZE);
215 end = round_down(ei->addr + ei->size, PAGE_SIZE);
218 if (ei->type == E820_RAM) {
219 if (end > end_pfn<<PAGE_SHIFT)
220 end_pfn = end>>PAGE_SHIFT;
222 if (end > end_pfn_map<<PAGE_SHIFT)
223 end_pfn_map = end>>PAGE_SHIFT;
227 if (end_pfn > end_pfn_map)
228 end_pfn_map = end_pfn;
229 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
230 end_pfn_map = MAXMEM>>PAGE_SHIFT;
231 if (end_pfn > end_user_pfn)
232 end_pfn = end_user_pfn;
233 if (end_pfn > end_pfn_map)
234 end_pfn = end_pfn_map;
240 * Compute how much memory is missing in a range.
241 * Unlike the other functions in this file the arguments are in page numbers.
244 e820_hole_size(unsigned long start_pfn, unsigned long end_pfn)
246 unsigned long ram = 0;
247 unsigned long start = start_pfn << PAGE_SHIFT;
248 unsigned long end = end_pfn << PAGE_SHIFT;
250 for (i = 0; i < e820.nr_map; i++) {
251 struct e820entry *ei = &e820.map[i];
252 unsigned long last, addr;
254 if (ei->type != E820_RAM ||
255 ei->addr+ei->size <= start ||
259 addr = round_up(ei->addr, PAGE_SIZE);
263 last = round_down(ei->addr + ei->size, PAGE_SIZE);
270 return ((end - start) - ram) >> PAGE_SHIFT;
274 * Mark e820 reserved areas as busy for the resource manager.
276 void __init e820_reserve_resources(struct e820entry *e820, int nr_map)
279 for (i = 0; i < nr_map; i++) {
280 struct resource *res;
281 res = alloc_bootmem_low(sizeof(struct resource));
282 switch (e820[i].type) {
283 case E820_RAM: res->name = "System RAM"; break;
284 case E820_ACPI: res->name = "ACPI Tables"; break;
285 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
286 default: res->name = "reserved";
288 res->start = e820[i].addr;
289 res->end = res->start + e820[i].size - 1;
290 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
291 request_resource(&iomem_resource, res);
292 if (e820[i].type == E820_RAM) {
294 * We don't know which RAM region contains kernel data,
295 * so we try it repeatedly and let the resource manager
298 request_resource(res, &code_resource);
299 request_resource(res, &data_resource);
301 request_resource(res, &crashk_res);
308 * Add a memory region to the kernel e820 map.
310 void __init add_memory_region(unsigned long start, unsigned long size, int type)
315 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
319 e820.map[x].addr = start;
320 e820.map[x].size = size;
321 e820.map[x].type = type;
325 void __init e820_print_map(char *who)
329 for (i = 0; i < e820.nr_map; i++) {
330 printk(" %s: %016Lx - %016Lx ", who,
331 (unsigned long long) e820.map[i].addr,
332 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
333 switch (e820.map[i].type) {
334 case E820_RAM: printk("(usable)\n");
337 printk("(reserved)\n");
340 printk("(ACPI data)\n");
343 printk("(ACPI NVS)\n");
345 default: printk("type %u\n", e820.map[i].type);
352 * Sanitize the BIOS e820 map.
354 * Some e820 responses include overlapping entries. The following
355 * replaces the original e820 map with a new one, removing overlaps.
358 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
360 struct change_member {
361 struct e820entry *pbios; /* pointer to original bios entry */
362 unsigned long long addr; /* address for this change point */
364 static struct change_member change_point_list[2*E820MAX] __initdata;
365 static struct change_member *change_point[2*E820MAX] __initdata;
366 static struct e820entry *overlap_list[E820MAX] __initdata;
367 static struct e820entry new_bios[E820MAX] __initdata;
368 struct change_member *change_tmp;
369 unsigned long current_type, last_type;
370 unsigned long long last_addr;
371 int chgidx, still_changing;
374 int old_nr, new_nr, chg_nr;
378 Visually we're performing the following (1,2,3,4 = memory types)...
380 Sample memory map (w/overlaps):
381 ____22__________________
382 ______________________4_
383 ____1111________________
384 _44_____________________
385 11111111________________
386 ____________________33__
387 ___________44___________
388 __________33333_________
389 ______________22________
390 ___________________2222_
391 _________111111111______
392 _____________________11_
393 _________________4______
395 Sanitized equivalent (no overlap):
396 1_______________________
397 _44_____________________
398 ___1____________________
399 ____22__________________
400 ______11________________
401 _________1______________
402 __________3_____________
403 ___________44___________
404 _____________33_________
405 _______________2________
406 ________________1_______
407 _________________4______
408 ___________________2____
409 ____________________33__
410 ______________________4_
413 /* if there's only one memory region, don't bother */
419 /* bail out if we find any unreasonable addresses in bios map */
420 for (i=0; i<old_nr; i++)
421 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
424 /* create pointers for initial change-point information (for sorting) */
425 for (i=0; i < 2*old_nr; i++)
426 change_point[i] = &change_point_list[i];
428 /* record all known change-points (starting and ending addresses),
429 omitting those that are for empty memory regions */
431 for (i=0; i < old_nr; i++) {
432 if (biosmap[i].size != 0) {
433 change_point[chgidx]->addr = biosmap[i].addr;
434 change_point[chgidx++]->pbios = &biosmap[i];
435 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
436 change_point[chgidx++]->pbios = &biosmap[i];
441 /* sort change-point list by memory addresses (low -> high) */
443 while (still_changing) {
445 for (i=1; i < chg_nr; i++) {
446 /* if <current_addr> > <last_addr>, swap */
447 /* or, if current=<start_addr> & last=<end_addr>, swap */
448 if ((change_point[i]->addr < change_point[i-1]->addr) ||
449 ((change_point[i]->addr == change_point[i-1]->addr) &&
450 (change_point[i]->addr == change_point[i]->pbios->addr) &&
451 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
454 change_tmp = change_point[i];
455 change_point[i] = change_point[i-1];
456 change_point[i-1] = change_tmp;
462 /* create a new bios memory map, removing overlaps */
463 overlap_entries=0; /* number of entries in the overlap table */
464 new_bios_entry=0; /* index for creating new bios map entries */
465 last_type = 0; /* start with undefined memory type */
466 last_addr = 0; /* start with 0 as last starting address */
467 /* loop through change-points, determining affect on the new bios map */
468 for (chgidx=0; chgidx < chg_nr; chgidx++)
470 /* keep track of all overlapping bios entries */
471 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
473 /* add map entry to overlap list (> 1 entry implies an overlap) */
474 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
478 /* remove entry from list (order independent, so swap with last) */
479 for (i=0; i<overlap_entries; i++)
481 if (overlap_list[i] == change_point[chgidx]->pbios)
482 overlap_list[i] = overlap_list[overlap_entries-1];
486 /* if there are overlapping entries, decide which "type" to use */
487 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
489 for (i=0; i<overlap_entries; i++)
490 if (overlap_list[i]->type > current_type)
491 current_type = overlap_list[i]->type;
492 /* continue building up new bios map based on this information */
493 if (current_type != last_type) {
494 if (last_type != 0) {
495 new_bios[new_bios_entry].size =
496 change_point[chgidx]->addr - last_addr;
497 /* move forward only if the new size was non-zero */
498 if (new_bios[new_bios_entry].size != 0)
499 if (++new_bios_entry >= E820MAX)
500 break; /* no more space left for new bios entries */
502 if (current_type != 0) {
503 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
504 new_bios[new_bios_entry].type = current_type;
505 last_addr=change_point[chgidx]->addr;
507 last_type = current_type;
510 new_nr = new_bios_entry; /* retain count for new bios entries */
512 /* copy new bios mapping into original location */
513 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
520 * Copy the BIOS e820 map into a safe place.
522 * Sanity-check it while we're at it..
524 * If we're lucky and live on a modern system, the setup code
525 * will have given us a memory map that we can use to properly
526 * set up memory. If we aren't, we'll fake a memory map.
528 * We check to see that the memory map contains at least 2 elements
529 * before we'll use it, because the detection code in setup.S may
530 * not be perfect and most every PC known to man has two memory
531 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
532 * thinkpad 560x, for example, does not cooperate with the memory
535 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
538 /* Only one memory region (or negative)? Ignore it */
546 unsigned long start = biosmap->addr;
547 unsigned long size = biosmap->size;
548 unsigned long end = start + size;
549 unsigned long type = biosmap->type;
551 /* Overflow in 64 bits? Ignore the memory map. */
557 * Some BIOSes claim RAM in the 640k - 1M region.
558 * Not right. Fix it up.
560 * This should be removed on Hammer which is supposed to not
561 * have non e820 covered ISA mappings there, but I had some strange
562 * problems so it stays for now. -AK
564 if (type == E820_RAM) {
565 if (start < 0x100000ULL && end > 0xA0000ULL) {
566 if (start < 0xA0000ULL)
567 add_memory_region(start, 0xA0000ULL-start, type);
568 if (end <= 0x100000ULL)
576 add_memory_region(start, size, type);
577 } while (biosmap++,--nr_map);
582 void __init setup_memory_region(void)
584 char *who = "BIOS-e820";
587 * Try to copy the BIOS-supplied E820-map.
589 * Otherwise fake a memory map; one section from 0k->640k,
590 * the next section from 1mb->appropriate_mem_k
592 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
593 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
594 unsigned long mem_size;
596 /* compare results from other methods and take the greater */
597 if (ALT_MEM_K < EXT_MEM_K) {
598 mem_size = EXT_MEM_K;
601 mem_size = ALT_MEM_K;
606 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
607 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
609 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
613 #else /* CONFIG_XEN */
615 void __init setup_memory_region(void)
618 struct xen_memory_map memmap;
620 * This is rather large for a stack variable but this early in
621 * the boot process we know we have plenty slack space.
623 struct e820entry map[E820MAX];
625 memmap.nr_entries = E820MAX;
626 set_xen_guest_handle(memmap.buffer, map);
628 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
629 if ( rc == -ENOSYS ) {
630 memmap.nr_entries = 1;
632 map[0].size = xen_start_info->nr_pages << PAGE_SHIFT;
633 /* 8MB slack (to balance backend allocations). */
634 map[0].size += 8 << 20;
635 map[0].type = E820_RAM;
640 sanitize_e820_map(map, (char *)&memmap.nr_entries);
642 BUG_ON(copy_e820_map(map, (char)memmap.nr_entries) < 0);
644 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
645 e820_print_map("Xen");
649 void __init parse_memopt(char *p, char **from)
652 unsigned long current_end;
655 end_user_pfn = memparse(p, from);
656 end_user_pfn >>= PAGE_SHIFT;
658 end = end_user_pfn<<PAGE_SHIFT;
660 current_end = e820.map[i].addr + e820.map[i].size;
662 if (current_end < end) {
664 * The e820 map ends before our requested size so
665 * extend the final entry to the requested address.
667 if (e820.map[i].type == E820_RAM)
668 e820.map[i].size = end - e820.map[i].addr;
670 add_memory_region(current_end, end - current_end, E820_RAM);
674 void __init parse_memmapopt(char *p, char **from)
676 unsigned long long start_at, mem_size;
678 mem_size = memparse(p, from);
681 start_at = memparse(p+1, from);
682 add_memory_region(start_at, mem_size, E820_RAM);
683 } else if (*p == '#') {
684 start_at = memparse(p+1, from);
685 add_memory_region(start_at, mem_size, E820_ACPI);
686 } else if (*p == '$') {
687 start_at = memparse(p+1, from);
688 add_memory_region(start_at, mem_size, E820_RESERVED);
690 end_user_pfn = (mem_size >> PAGE_SHIFT);
695 unsigned long pci_mem_start = 0xaeedbabe;
698 * Search for the biggest gap in the low 32 bits of the e820
699 * memory space. We pass this space to PCI to assign MMIO resources
700 * for hotplug or unconfigured devices in.
701 * Hopefully the BIOS let enough space left.
703 __init void e820_setup_gap(struct e820entry *e820, int nr_map)
705 unsigned long gapstart, gapsize, round;
710 last = 0x100000000ull;
711 gapstart = 0x10000000;
715 unsigned long long start = e820[i].addr;
716 unsigned long long end = start + e820[i].size;
719 * Since "last" is at most 4GB, we know we'll
720 * fit in 32 bits if this condition is true
723 unsigned long gap = last - end;
736 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
737 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
738 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
742 * See how much we want to round up: start off with
743 * rounding to the next 1MB area.
746 while ((gapsize >> 4) > round)
748 /* Fun with two's complement */
749 pci_mem_start = (gapstart + round) & -round;
751 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
752 pci_mem_start, gapstart, gapsize);