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>
26 #include <xen/interface/memory.h>
29 * PFN of last memory page.
31 unsigned long end_pfn;
32 EXPORT_SYMBOL(end_pfn);
35 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
36 * The direct mapping extends to end_pfn_map, so that we can directly access
37 * apertures, ACPI and other tables without having to play with fixmaps.
39 unsigned long end_pfn_map;
42 * Last pfn which the user wants to use.
44 unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT;
46 extern struct resource code_resource, data_resource;
48 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
49 static inline int bad_addr(unsigned long *addrp, unsigned long size)
51 unsigned long addr = *addrp, last = addr + size;
54 /* various gunk below that needed for SMP startup */
60 /* direct mapping tables of the kernel */
61 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
62 *addrp = table_end << PAGE_SHIFT;
67 #ifdef CONFIG_BLK_DEV_INITRD
68 if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
69 addr < INITRD_START+INITRD_SIZE) {
70 *addrp = INITRD_START + INITRD_SIZE;
74 /* kernel code + 640k memory hole (later should not be needed, but
75 be paranoid for now) */
76 if (last >= 640*1024 && addr < 1024*1024) {
80 if (last >= __pa_symbol(&_text) && last < __pa_symbol(&_end)) {
81 *addrp = __pa_symbol(&_end);
85 if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
86 *addrp = ebda_addr + ebda_size;
90 /* XXX ramdisk image here? */
92 if (last < (table_end<<PAGE_SHIFT)) {
93 *addrp = table_end << PAGE_SHIFT;
101 * This function checks if any part of the range <start,end> is mapped
105 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
108 for (i = 0; i < e820.nr_map; i++) {
109 struct e820entry *ei = &e820.map[i];
110 if (type && ei->type != type)
112 if (ei->addr >= end || ei->addr + ei->size <= start)
120 * This function checks if the entire range <start,end> is mapped with type.
122 * Note: this function only works correct if the e820 table is sorted and
123 * not-overlapping, which is the case
125 int __init e820_all_mapped(unsigned long start, unsigned long end, unsigned type)
128 for (i = 0; i < e820.nr_map; i++) {
129 struct e820entry *ei = &e820.map[i];
130 if (type && ei->type != type)
132 /* is the region (part) in overlap with the current region ?*/
133 if (ei->addr >= end || ei->addr + ei->size <= start)
136 /* if the region is at the beginning of <start,end> we move
137 * start to the end of the region since it's ok until there
139 if (ei->addr <= start)
140 start = ei->addr + ei->size;
141 /* if start is now at or beyond end, we're done, full coverage */
143 return 1; /* we're done */
149 * Find a free area in a specific range.
151 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
154 for (i = 0; i < e820.nr_map; i++) {
155 struct e820entry *ei = &e820.map[i];
156 unsigned long addr = ei->addr, last;
157 if (ei->type != E820_RAM)
161 if (addr > ei->addr + ei->size)
163 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
166 if (last > ei->addr + ei->size)
176 * Free bootmem based on the e820 table for a node.
178 void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end)
181 for (i = 0; i < e820.nr_map; i++) {
182 struct e820entry *ei = &e820.map[i];
183 unsigned long last, addr;
185 if (ei->type != E820_RAM ||
186 ei->addr+ei->size <= start ||
190 addr = round_up(ei->addr, PAGE_SIZE);
194 last = round_down(ei->addr + ei->size, PAGE_SIZE);
198 if (last > addr && last-addr >= PAGE_SIZE)
199 free_bootmem_node(pgdat, addr, last-addr);
204 * Find the highest page frame number we have available
206 unsigned long __init e820_end_of_ram(void)
209 unsigned long end_pfn = 0;
211 for (i = 0; i < e820.nr_map; i++) {
212 struct e820entry *ei = &e820.map[i];
213 unsigned long start, end;
215 start = round_up(ei->addr, PAGE_SIZE);
216 end = round_down(ei->addr + ei->size, PAGE_SIZE);
219 if (ei->type == E820_RAM) {
220 if (end > end_pfn<<PAGE_SHIFT)
221 end_pfn = end>>PAGE_SHIFT;
223 if (end > end_pfn_map<<PAGE_SHIFT)
224 end_pfn_map = end>>PAGE_SHIFT;
228 if (end_pfn > end_pfn_map)
229 end_pfn_map = end_pfn;
230 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
231 end_pfn_map = MAXMEM>>PAGE_SHIFT;
232 if (end_pfn > end_user_pfn)
233 end_pfn = end_user_pfn;
234 if (end_pfn > end_pfn_map)
235 end_pfn = end_pfn_map;
241 * Compute how much memory is missing in a range.
242 * Unlike the other functions in this file the arguments are in page numbers.
245 e820_hole_size(unsigned long start_pfn, unsigned long end_pfn)
247 unsigned long ram = 0;
248 unsigned long start = start_pfn << PAGE_SHIFT;
249 unsigned long end = end_pfn << PAGE_SHIFT;
251 for (i = 0; i < e820.nr_map; i++) {
252 struct e820entry *ei = &e820.map[i];
253 unsigned long last, addr;
255 if (ei->type != E820_RAM ||
256 ei->addr+ei->size <= start ||
260 addr = round_up(ei->addr, PAGE_SIZE);
264 last = round_down(ei->addr + ei->size, PAGE_SIZE);
271 return ((end - start) - ram) >> PAGE_SHIFT;
275 * Mark e820 reserved areas as busy for the resource manager.
277 void __init e820_reserve_resources(struct e820entry *e820, int nr_map)
280 for (i = 0; i < nr_map; i++) {
281 struct resource *res;
282 res = alloc_bootmem_low(sizeof(struct resource));
283 switch (e820[i].type) {
284 case E820_RAM: res->name = "System RAM"; break;
285 case E820_ACPI: res->name = "ACPI Tables"; break;
286 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
287 default: res->name = "reserved";
289 res->start = e820[i].addr;
290 res->end = res->start + e820[i].size - 1;
291 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
292 request_resource(&iomem_resource, res);
293 if (e820[i].type == E820_RAM) {
295 * We don't know which RAM region contains kernel data,
296 * so we try it repeatedly and let the resource manager
300 request_resource(res, &code_resource);
301 request_resource(res, &data_resource);
304 request_resource(res, &crashk_res);
311 * Add a memory region to the kernel e820 map.
313 void __init add_memory_region(unsigned long start, unsigned long size, int type)
318 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
322 e820.map[x].addr = start;
323 e820.map[x].size = size;
324 e820.map[x].type = type;
328 void __init e820_print_map(char *who)
332 for (i = 0; i < e820.nr_map; i++) {
333 printk(" %s: %016Lx - %016Lx ", who,
334 (unsigned long long) e820.map[i].addr,
335 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
336 switch (e820.map[i].type) {
337 case E820_RAM: printk("(usable)\n");
340 printk("(reserved)\n");
343 printk("(ACPI data)\n");
346 printk("(ACPI NVS)\n");
348 default: printk("type %u\n", e820.map[i].type);
355 * Sanitize the BIOS e820 map.
357 * Some e820 responses include overlapping entries. The following
358 * replaces the original e820 map with a new one, removing overlaps.
361 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
363 struct change_member {
364 struct e820entry *pbios; /* pointer to original bios entry */
365 unsigned long long addr; /* address for this change point */
367 static struct change_member change_point_list[2*E820MAX] __initdata;
368 static struct change_member *change_point[2*E820MAX] __initdata;
369 static struct e820entry *overlap_list[E820MAX] __initdata;
370 static struct e820entry new_bios[E820MAX] __initdata;
371 struct change_member *change_tmp;
372 unsigned long current_type, last_type;
373 unsigned long long last_addr;
374 int chgidx, still_changing;
377 int old_nr, new_nr, chg_nr;
381 Visually we're performing the following (1,2,3,4 = memory types)...
383 Sample memory map (w/overlaps):
384 ____22__________________
385 ______________________4_
386 ____1111________________
387 _44_____________________
388 11111111________________
389 ____________________33__
390 ___________44___________
391 __________33333_________
392 ______________22________
393 ___________________2222_
394 _________111111111______
395 _____________________11_
396 _________________4______
398 Sanitized equivalent (no overlap):
399 1_______________________
400 _44_____________________
401 ___1____________________
402 ____22__________________
403 ______11________________
404 _________1______________
405 __________3_____________
406 ___________44___________
407 _____________33_________
408 _______________2________
409 ________________1_______
410 _________________4______
411 ___________________2____
412 ____________________33__
413 ______________________4_
416 /* if there's only one memory region, don't bother */
422 /* bail out if we find any unreasonable addresses in bios map */
423 for (i=0; i<old_nr; i++)
424 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
427 /* create pointers for initial change-point information (for sorting) */
428 for (i=0; i < 2*old_nr; i++)
429 change_point[i] = &change_point_list[i];
431 /* record all known change-points (starting and ending addresses),
432 omitting those that are for empty memory regions */
434 for (i=0; i < old_nr; i++) {
435 if (biosmap[i].size != 0) {
436 change_point[chgidx]->addr = biosmap[i].addr;
437 change_point[chgidx++]->pbios = &biosmap[i];
438 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
439 change_point[chgidx++]->pbios = &biosmap[i];
444 /* sort change-point list by memory addresses (low -> high) */
446 while (still_changing) {
448 for (i=1; i < chg_nr; i++) {
449 /* if <current_addr> > <last_addr>, swap */
450 /* or, if current=<start_addr> & last=<end_addr>, swap */
451 if ((change_point[i]->addr < change_point[i-1]->addr) ||
452 ((change_point[i]->addr == change_point[i-1]->addr) &&
453 (change_point[i]->addr == change_point[i]->pbios->addr) &&
454 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
457 change_tmp = change_point[i];
458 change_point[i] = change_point[i-1];
459 change_point[i-1] = change_tmp;
465 /* create a new bios memory map, removing overlaps */
466 overlap_entries=0; /* number of entries in the overlap table */
467 new_bios_entry=0; /* index for creating new bios map entries */
468 last_type = 0; /* start with undefined memory type */
469 last_addr = 0; /* start with 0 as last starting address */
470 /* loop through change-points, determining affect on the new bios map */
471 for (chgidx=0; chgidx < chg_nr; chgidx++)
473 /* keep track of all overlapping bios entries */
474 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
476 /* add map entry to overlap list (> 1 entry implies an overlap) */
477 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
481 /* remove entry from list (order independent, so swap with last) */
482 for (i=0; i<overlap_entries; i++)
484 if (overlap_list[i] == change_point[chgidx]->pbios)
485 overlap_list[i] = overlap_list[overlap_entries-1];
489 /* if there are overlapping entries, decide which "type" to use */
490 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
492 for (i=0; i<overlap_entries; i++)
493 if (overlap_list[i]->type > current_type)
494 current_type = overlap_list[i]->type;
495 /* continue building up new bios map based on this information */
496 if (current_type != last_type) {
497 if (last_type != 0) {
498 new_bios[new_bios_entry].size =
499 change_point[chgidx]->addr - last_addr;
500 /* move forward only if the new size was non-zero */
501 if (new_bios[new_bios_entry].size != 0)
502 if (++new_bios_entry >= E820MAX)
503 break; /* no more space left for new bios entries */
505 if (current_type != 0) {
506 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
507 new_bios[new_bios_entry].type = current_type;
508 last_addr=change_point[chgidx]->addr;
510 last_type = current_type;
513 new_nr = new_bios_entry; /* retain count for new bios entries */
515 /* copy new bios mapping into original location */
516 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
523 * Copy the BIOS e820 map into a safe place.
525 * Sanity-check it while we're at it..
527 * If we're lucky and live on a modern system, the setup code
528 * will have given us a memory map that we can use to properly
529 * set up memory. If we aren't, we'll fake a memory map.
531 * We check to see that the memory map contains at least 2 elements
532 * before we'll use it, because the detection code in setup.S may
533 * not be perfect and most every PC known to man has two memory
534 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
535 * thinkpad 560x, for example, does not cooperate with the memory
538 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
541 /* Only one memory region (or negative)? Ignore it */
549 unsigned long start = biosmap->addr;
550 unsigned long size = biosmap->size;
551 unsigned long end = start + size;
552 unsigned long type = biosmap->type;
554 /* Overflow in 64 bits? Ignore the memory map. */
560 * Some BIOSes claim RAM in the 640k - 1M region.
561 * Not right. Fix it up.
563 * This should be removed on Hammer which is supposed to not
564 * have non e820 covered ISA mappings there, but I had some strange
565 * problems so it stays for now. -AK
567 if (type == E820_RAM) {
568 if (start < 0x100000ULL && end > 0xA0000ULL) {
569 if (start < 0xA0000ULL)
570 add_memory_region(start, 0xA0000ULL-start, type);
571 if (end <= 0x100000ULL)
579 add_memory_region(start, size, type);
580 } while (biosmap++,--nr_map);
585 void __init setup_memory_region(void)
587 char *who = "BIOS-e820";
590 * Try to copy the BIOS-supplied E820-map.
592 * Otherwise fake a memory map; one section from 0k->640k,
593 * the next section from 1mb->appropriate_mem_k
595 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
596 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) {
597 unsigned long mem_size;
599 /* compare results from other methods and take the greater */
600 if (ALT_MEM_K < EXT_MEM_K) {
601 mem_size = EXT_MEM_K;
604 mem_size = ALT_MEM_K;
609 add_memory_region(0, LOWMEMSIZE(), E820_RAM);
610 add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
612 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
616 #else /* CONFIG_XEN */
618 void __init setup_memory_region(void)
621 struct xen_memory_map memmap;
623 * This is rather large for a stack variable but this early in
624 * the boot process we know we have plenty slack space.
626 struct e820entry map[E820MAX];
628 memmap.nr_entries = E820MAX;
629 set_xen_guest_handle(memmap.buffer, map);
631 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
632 if ( rc == -ENOSYS ) {
633 memmap.nr_entries = 1;
635 map[0].size = xen_start_info->nr_pages << PAGE_SHIFT;
636 /* 8MB slack (to balance backend allocations). */
637 map[0].size += 8 << 20;
638 map[0].type = E820_RAM;
643 sanitize_e820_map(map, (char *)&memmap.nr_entries);
645 BUG_ON(copy_e820_map(map, (char)memmap.nr_entries) < 0);
647 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
648 e820_print_map("Xen");
652 void __init parse_memopt(char *p, char **from)
655 unsigned long current_end;
658 end_user_pfn = memparse(p, from);
659 end_user_pfn >>= PAGE_SHIFT;
661 end = end_user_pfn<<PAGE_SHIFT;
663 current_end = e820.map[i].addr + e820.map[i].size;
665 if (current_end < end) {
667 * The e820 map ends before our requested size so
668 * extend the final entry to the requested address.
670 if (e820.map[i].type == E820_RAM)
671 e820.map[i].size = end - e820.map[i].addr;
673 add_memory_region(current_end, end - current_end, E820_RAM);
677 void __init parse_memmapopt(char *p, char **from)
679 unsigned long long start_at, mem_size;
681 mem_size = memparse(p, from);
684 start_at = memparse(p+1, from);
685 add_memory_region(start_at, mem_size, E820_RAM);
686 } else if (*p == '#') {
687 start_at = memparse(p+1, from);
688 add_memory_region(start_at, mem_size, E820_ACPI);
689 } else if (*p == '$') {
690 start_at = memparse(p+1, from);
691 add_memory_region(start_at, mem_size, E820_RESERVED);
693 end_user_pfn = (mem_size >> PAGE_SHIFT);
698 unsigned long pci_mem_start = 0xaeedbabe;
699 EXPORT_SYMBOL(pci_mem_start);
702 * Search for the biggest gap in the low 32 bits of the e820
703 * memory space. We pass this space to PCI to assign MMIO resources
704 * for hotplug or unconfigured devices in.
705 * Hopefully the BIOS let enough space left.
707 __init void e820_setup_gap(struct e820entry *e820, int nr_map)
709 unsigned long gapstart, gapsize, round;
714 last = 0x100000000ull;
715 gapstart = 0x10000000;
719 unsigned long long start = e820[i].addr;
720 unsigned long long end = start + e820[i].size;
723 * Since "last" is at most 4GB, we know we'll
724 * fit in 32 bits if this condition is true
727 unsigned long gap = last - end;
740 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
741 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
742 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
746 * See how much we want to round up: start off with
747 * rounding to the next 1MB area.
750 while ((gapsize >> 4) > round)
752 /* Fun with two's complement */
753 pci_mem_start = (gapstart + round) & -round;
755 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
756 pci_mem_start, gapstart, gapsize);