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 <linux/suspend.h>
21 #include <linux/pfn.h>
23 #include <asm/pgtable.h>
26 #include <asm/proto.h>
27 #include <asm/setup.h>
28 #include <asm/sections.h>
29 #include <asm/kdebug.h>
34 * PFN of last memory page.
36 unsigned long end_pfn;
39 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
40 * The direct mapping extends to end_pfn_map, so that we can directly access
41 * apertures, ACPI and other tables without having to play with fixmaps.
43 unsigned long end_pfn_map;
46 * Last pfn which the user wants to use.
48 static unsigned long __initdata end_user_pfn = MAXMEM>>PAGE_SHIFT;
51 * Early reserved memory areas.
53 #define MAX_EARLY_RES 20
56 unsigned long start, end;
59 static struct early_res early_res[MAX_EARLY_RES] __initdata = {
60 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
62 { SMP_TRAMPOLINE_BASE, SMP_TRAMPOLINE_BASE + 2*PAGE_SIZE, "SMP_TRAMPOLINE" },
67 void __init reserve_early(unsigned long start, unsigned long end, char *name)
71 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
73 if (end > r->start && start < r->end)
74 panic("Overlapping early reservations %lx-%lx %s to %lx-%lx %s\n",
75 start, end - 1, name?name:"", r->start, r->end - 1, r->name);
77 if (i >= MAX_EARLY_RES)
78 panic("Too many early reservations");
83 strncpy(r->name, name, sizeof(r->name) - 1);
86 void __init early_res_to_bootmem(void)
89 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
90 struct early_res *r = &early_res[i];
91 printk(KERN_INFO "early res: %d [%lx-%lx] %s\n", i,
92 r->start, r->end - 1, r->name);
93 reserve_bootmem_generic(r->start, r->end - r->start);
97 /* Check for already reserved areas */
99 bad_addr(unsigned long *addrp, unsigned long size, unsigned long align)
102 unsigned long addr = *addrp, last;
106 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
107 struct early_res *r = &early_res[i];
108 if (last >= r->start && addr < r->end) {
109 *addrp = addr = round_up(r->end, align);
118 * This function checks if any part of the range <start,end> is mapped
122 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
126 for (i = 0; i < e820.nr_map; i++) {
127 struct e820entry *ei = &e820.map[i];
129 if (type && ei->type != type)
131 if (ei->addr >= end || ei->addr + ei->size <= start)
137 EXPORT_SYMBOL_GPL(e820_any_mapped);
140 * This function checks if the entire range <start,end> is mapped with type.
142 * Note: this function only works correct if the e820 table is sorted and
143 * not-overlapping, which is the case
145 int __init e820_all_mapped(unsigned long start, unsigned long end,
150 for (i = 0; i < e820.nr_map; i++) {
151 struct e820entry *ei = &e820.map[i];
153 if (type && ei->type != type)
155 /* is the region (part) in overlap with the current region ?*/
156 if (ei->addr >= end || ei->addr + ei->size <= start)
159 /* if the region is at the beginning of <start,end> we move
160 * start to the end of the region since it's ok until there
162 if (ei->addr <= start)
163 start = ei->addr + ei->size;
165 * if start is now at or beyond end, we're done, full
175 * Find a free area with specified alignment in a specific range.
177 unsigned long __init find_e820_area(unsigned long start, unsigned long end,
178 unsigned long size, unsigned long align)
182 for (i = 0; i < e820.nr_map; i++) {
183 struct e820entry *ei = &e820.map[i];
184 unsigned long addr, last;
185 unsigned long ei_last;
187 if (ei->type != E820_RAM)
189 addr = round_up(ei->addr, align);
190 ei_last = ei->addr + ei->size;
192 addr = round_up(start, align);
195 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
208 * Find the highest page frame number we have available
210 unsigned long __init e820_end_of_ram(void)
212 unsigned long end_pfn;
214 end_pfn = find_max_pfn_with_active_regions();
216 if (end_pfn > end_pfn_map)
217 end_pfn_map = end_pfn;
218 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
219 end_pfn_map = MAXMEM>>PAGE_SHIFT;
220 if (end_pfn > end_user_pfn)
221 end_pfn = end_user_pfn;
222 if (end_pfn > end_pfn_map)
223 end_pfn = end_pfn_map;
225 printk(KERN_INFO "end_pfn_map = %lu\n", end_pfn_map);
230 * Mark e820 reserved areas as busy for the resource manager.
232 void __init e820_reserve_resources(struct resource *code_resource,
233 struct resource *data_resource, struct resource *bss_resource)
236 for (i = 0; i < e820.nr_map; i++) {
237 struct resource *res;
238 res = alloc_bootmem_low(sizeof(struct resource));
239 switch (e820.map[i].type) {
240 case E820_RAM: res->name = "System RAM"; break;
241 case E820_ACPI: res->name = "ACPI Tables"; break;
242 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
243 default: res->name = "reserved";
245 res->start = e820.map[i].addr;
246 res->end = res->start + e820.map[i].size - 1;
247 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
248 request_resource(&iomem_resource, res);
249 if (e820.map[i].type == E820_RAM) {
251 * We don't know which RAM region contains kernel data,
252 * so we try it repeatedly and let the resource manager
255 request_resource(res, code_resource);
256 request_resource(res, data_resource);
257 request_resource(res, bss_resource);
259 if (crashk_res.start != crashk_res.end)
260 request_resource(res, &crashk_res);
267 * Find the ranges of physical addresses that do not correspond to
268 * e820 RAM areas and mark the corresponding pages as nosave for software
269 * suspend and suspend to RAM.
271 * This function requires the e820 map to be sorted and without any
272 * overlapping entries and assumes the first e820 area to be RAM.
274 void __init e820_mark_nosave_regions(void)
279 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
280 for (i = 1; i < e820.nr_map; i++) {
281 struct e820entry *ei = &e820.map[i];
283 if (paddr < ei->addr)
284 register_nosave_region(PFN_DOWN(paddr),
287 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
288 if (ei->type != E820_RAM)
289 register_nosave_region(PFN_UP(ei->addr),
292 if (paddr >= (end_pfn << PAGE_SHIFT))
298 * Finds an active region in the address range from start_pfn to end_pfn and
299 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
301 static int __init e820_find_active_region(const struct e820entry *ei,
302 unsigned long start_pfn,
303 unsigned long end_pfn,
304 unsigned long *ei_startpfn,
305 unsigned long *ei_endpfn)
307 *ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
308 *ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE) >> PAGE_SHIFT;
310 /* Skip map entries smaller than a page */
311 if (*ei_startpfn >= *ei_endpfn)
314 /* Check if end_pfn_map should be updated */
315 if (ei->type != E820_RAM && *ei_endpfn > end_pfn_map)
316 end_pfn_map = *ei_endpfn;
318 /* Skip if map is outside the node */
319 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
320 *ei_startpfn >= end_pfn)
323 /* Check for overlaps */
324 if (*ei_startpfn < start_pfn)
325 *ei_startpfn = start_pfn;
326 if (*ei_endpfn > end_pfn)
327 *ei_endpfn = end_pfn;
329 /* Obey end_user_pfn to save on memmap */
330 if (*ei_startpfn >= end_user_pfn)
332 if (*ei_endpfn > end_user_pfn)
333 *ei_endpfn = end_user_pfn;
338 /* Walk the e820 map and register active regions within a node */
340 e820_register_active_regions(int nid, unsigned long start_pfn,
341 unsigned long end_pfn)
343 unsigned long ei_startpfn;
344 unsigned long ei_endpfn;
347 for (i = 0; i < e820.nr_map; i++)
348 if (e820_find_active_region(&e820.map[i],
350 &ei_startpfn, &ei_endpfn))
351 add_active_range(nid, ei_startpfn, ei_endpfn);
355 * Add a memory region to the kernel e820 map.
357 void __init add_memory_region(unsigned long start, unsigned long size, int type)
362 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
366 e820.map[x].addr = start;
367 e820.map[x].size = size;
368 e820.map[x].type = type;
373 * Find the hole size (in bytes) in the memory range.
374 * @start: starting address of the memory range to scan
375 * @end: ending address of the memory range to scan
377 unsigned long __init e820_hole_size(unsigned long start, unsigned long end)
379 unsigned long start_pfn = start >> PAGE_SHIFT;
380 unsigned long end_pfn = end >> PAGE_SHIFT;
381 unsigned long ei_startpfn, ei_endpfn, ram = 0;
384 for (i = 0; i < e820.nr_map; i++) {
385 if (e820_find_active_region(&e820.map[i],
387 &ei_startpfn, &ei_endpfn))
388 ram += ei_endpfn - ei_startpfn;
390 return end - start - (ram << PAGE_SHIFT);
393 static void __init e820_print_map(char *who)
397 for (i = 0; i < e820.nr_map; i++) {
398 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
399 (unsigned long long) e820.map[i].addr,
401 (e820.map[i].addr + e820.map[i].size));
402 switch (e820.map[i].type) {
404 printk(KERN_CONT "(usable)\n");
407 printk(KERN_CONT "(reserved)\n");
410 printk(KERN_CONT "(ACPI data)\n");
413 printk(KERN_CONT "(ACPI NVS)\n");
416 printk(KERN_CONT "type %u\n", e820.map[i].type);
423 * Sanitize the BIOS e820 map.
425 * Some e820 responses include overlapping entries. The following
426 * replaces the original e820 map with a new one, removing overlaps.
429 static int __init sanitize_e820_map(struct e820entry *biosmap, char *pnr_map)
431 struct change_member {
432 struct e820entry *pbios; /* pointer to original bios entry */
433 unsigned long long addr; /* address for this change point */
435 static struct change_member change_point_list[2*E820MAX] __initdata;
436 static struct change_member *change_point[2*E820MAX] __initdata;
437 static struct e820entry *overlap_list[E820MAX] __initdata;
438 static struct e820entry new_bios[E820MAX] __initdata;
439 struct change_member *change_tmp;
440 unsigned long current_type, last_type;
441 unsigned long long last_addr;
442 int chgidx, still_changing;
445 int old_nr, new_nr, chg_nr;
449 Visually we're performing the following
450 (1,2,3,4 = memory types)...
452 Sample memory map (w/overlaps):
453 ____22__________________
454 ______________________4_
455 ____1111________________
456 _44_____________________
457 11111111________________
458 ____________________33__
459 ___________44___________
460 __________33333_________
461 ______________22________
462 ___________________2222_
463 _________111111111______
464 _____________________11_
465 _________________4______
467 Sanitized equivalent (no overlap):
468 1_______________________
469 _44_____________________
470 ___1____________________
471 ____22__________________
472 ______11________________
473 _________1______________
474 __________3_____________
475 ___________44___________
476 _____________33_________
477 _______________2________
478 ________________1_______
479 _________________4______
480 ___________________2____
481 ____________________33__
482 ______________________4_
485 /* if there's only one memory region, don't bother */
491 /* bail out if we find any unreasonable addresses in bios map */
492 for (i = 0; i < old_nr; i++)
493 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
496 /* create pointers for initial change-point information (for sorting) */
497 for (i = 0; i < 2 * old_nr; i++)
498 change_point[i] = &change_point_list[i];
500 /* record all known change-points (starting and ending addresses),
501 omitting those that are for empty memory regions */
503 for (i = 0; i < old_nr; i++) {
504 if (biosmap[i].size != 0) {
505 change_point[chgidx]->addr = biosmap[i].addr;
506 change_point[chgidx++]->pbios = &biosmap[i];
507 change_point[chgidx]->addr = biosmap[i].addr +
509 change_point[chgidx++]->pbios = &biosmap[i];
514 /* sort change-point list by memory addresses (low -> high) */
516 while (still_changing) {
518 for (i = 1; i < chg_nr; i++) {
519 unsigned long long curaddr, lastaddr;
520 unsigned long long curpbaddr, lastpbaddr;
522 curaddr = change_point[i]->addr;
523 lastaddr = change_point[i - 1]->addr;
524 curpbaddr = change_point[i]->pbios->addr;
525 lastpbaddr = change_point[i - 1]->pbios->addr;
528 * swap entries, when:
530 * curaddr > lastaddr or
531 * curaddr == lastaddr and curaddr == curpbaddr and
532 * lastaddr != lastpbaddr
534 if (curaddr < lastaddr ||
535 (curaddr == lastaddr && curaddr == curpbaddr &&
536 lastaddr != lastpbaddr)) {
537 change_tmp = change_point[i];
538 change_point[i] = change_point[i-1];
539 change_point[i-1] = change_tmp;
545 /* create a new bios memory map, removing overlaps */
546 overlap_entries = 0; /* number of entries in the overlap table */
547 new_bios_entry = 0; /* index for creating new bios map entries */
548 last_type = 0; /* start with undefined memory type */
549 last_addr = 0; /* start with 0 as last starting address */
551 /* loop through change-points, determining affect on the new bios map */
552 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
553 /* keep track of all overlapping bios entries */
554 if (change_point[chgidx]->addr ==
555 change_point[chgidx]->pbios->addr) {
557 * add map entry to overlap list (> 1 entry
558 * implies an overlap)
560 overlap_list[overlap_entries++] =
561 change_point[chgidx]->pbios;
564 * remove entry from list (order independent,
567 for (i = 0; i < overlap_entries; i++) {
568 if (overlap_list[i] ==
569 change_point[chgidx]->pbios)
571 overlap_list[overlap_entries-1];
576 * if there are overlapping entries, decide which
577 * "type" to use (larger value takes precedence --
578 * 1=usable, 2,3,4,4+=unusable)
581 for (i = 0; i < overlap_entries; i++)
582 if (overlap_list[i]->type > current_type)
583 current_type = overlap_list[i]->type;
585 * continue building up new bios map based on this
588 if (current_type != last_type) {
589 if (last_type != 0) {
590 new_bios[new_bios_entry].size =
591 change_point[chgidx]->addr - last_addr;
593 * move forward only if the new size
596 if (new_bios[new_bios_entry].size != 0)
598 * no more space left for new
601 if (++new_bios_entry >= E820MAX)
604 if (current_type != 0) {
605 new_bios[new_bios_entry].addr =
606 change_point[chgidx]->addr;
607 new_bios[new_bios_entry].type = current_type;
608 last_addr = change_point[chgidx]->addr;
610 last_type = current_type;
613 /* retain count for new bios entries */
614 new_nr = new_bios_entry;
616 /* copy new bios mapping into original location */
617 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
624 * Copy the BIOS e820 map into a safe place.
626 * Sanity-check it while we're at it..
628 * If we're lucky and live on a modern system, the setup code
629 * will have given us a memory map that we can use to properly
630 * set up memory. If we aren't, we'll fake a memory map.
632 static int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
634 /* Only one memory region (or negative)? Ignore it */
639 unsigned long start = biosmap->addr;
640 unsigned long size = biosmap->size;
641 unsigned long end = start + size;
642 unsigned long type = biosmap->type;
644 /* Overflow in 64 bits? Ignore the memory map. */
648 add_memory_region(start, size, type);
649 } while (biosmap++, --nr_map);
653 static void early_panic(char *msg)
659 /* We're not void only for x86 32-bit compat */
660 char * __init machine_specific_memory_setup(void)
662 char *who = "BIOS-e820";
664 * Try to copy the BIOS-supplied E820-map.
666 * Otherwise fake a memory map; one section from 0k->640k,
667 * the next section from 1mb->appropriate_mem_k
669 sanitize_e820_map(boot_params.e820_map, &boot_params.e820_entries);
670 if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0)
671 early_panic("Cannot find a valid memory map");
672 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
675 /* In case someone cares... */
679 static int __init parse_memopt(char *p)
683 end_user_pfn = memparse(p, &p);
684 end_user_pfn >>= PAGE_SHIFT;
687 early_param("mem", parse_memopt);
689 static int userdef __initdata;
691 static int __init parse_memmap_opt(char *p)
694 unsigned long long start_at, mem_size;
696 if (!strcmp(p, "exactmap")) {
697 #ifdef CONFIG_CRASH_DUMP
699 * If we are doing a crash dump, we still need to know
700 * the real mem size before original memory map is
703 e820_register_active_regions(0, 0, -1UL);
704 saved_max_pfn = e820_end_of_ram();
705 remove_all_active_ranges();
714 mem_size = memparse(p, &p);
720 start_at = memparse(p+1, &p);
721 add_memory_region(start_at, mem_size, E820_RAM);
722 } else if (*p == '#') {
723 start_at = memparse(p+1, &p);
724 add_memory_region(start_at, mem_size, E820_ACPI);
725 } else if (*p == '$') {
726 start_at = memparse(p+1, &p);
727 add_memory_region(start_at, mem_size, E820_RESERVED);
729 end_user_pfn = (mem_size >> PAGE_SHIFT);
731 return *p == '\0' ? 0 : -EINVAL;
733 early_param("memmap", parse_memmap_opt);
735 void __init finish_e820_parsing(void)
738 char nr = e820.nr_map;
740 if (sanitize_e820_map(e820.map, &nr) < 0)
741 early_panic("Invalid user supplied memory map");
744 printk(KERN_INFO "user-defined physical RAM map:\n");
745 e820_print_map("user");
749 void __init update_memory_range(u64 start, u64 size, unsigned old_type,
754 BUG_ON(old_type == new_type);
756 for (i = 0; i < e820.nr_map; i++) {
757 struct e820entry *ei = &e820.map[i];
758 u64 final_start, final_end;
759 if (ei->type != old_type)
761 /* totally covered? */
762 if (ei->addr >= start && ei->size <= size) {
766 /* partially covered */
767 final_start = max(start, ei->addr);
768 final_end = min(start + size, ei->addr + ei->size);
769 if (final_start >= final_end)
771 add_memory_region(final_start, final_end - final_start,
776 void __init update_e820(void)
780 nr_map = e820.nr_map;
781 if (sanitize_e820_map(e820.map, &nr_map))
783 e820.nr_map = nr_map;
784 printk(KERN_INFO "modified physical RAM map:\n");
785 e820_print_map("modified");
788 unsigned long pci_mem_start = 0xaeedbabe;
789 EXPORT_SYMBOL(pci_mem_start);
792 * Search for the biggest gap in the low 32 bits of the e820
793 * memory space. We pass this space to PCI to assign MMIO resources
794 * for hotplug or unconfigured devices in.
795 * Hopefully the BIOS let enough space left.
797 __init void e820_setup_gap(void)
799 unsigned long gapstart, gapsize, round;
804 last = 0x100000000ull;
805 gapstart = 0x10000000;
809 unsigned long long start = e820.map[i].addr;
810 unsigned long long end = start + e820.map[i].size;
813 * Since "last" is at most 4GB, we know we'll
814 * fit in 32 bits if this condition is true
817 unsigned long gap = last - end;
830 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
831 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
833 KERN_ERR "PCI: Unassigned devices with 32bit resource "
834 "registers may break!\n");
838 * See how much we want to round up: start off with
839 * rounding to the next 1MB area.
842 while ((gapsize >> 4) > round)
844 /* Fun with two's complement */
845 pci_mem_start = (gapstart + round) & -round;
848 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
849 pci_mem_start, gapstart, gapsize);
852 int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
856 if (slot < 0 || slot >= e820.nr_map)
858 for (i = slot; i < e820.nr_map; i++) {
859 if (e820.map[i].type != E820_RAM)
863 if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
865 *addr = e820.map[i].addr;
866 *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
867 max_pfn << PAGE_SHIFT) - *addr;