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 */
98 static inline int bad_addr(unsigned long *addrp, unsigned long size)
101 unsigned long addr = *addrp, last;
105 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
106 struct early_res *r = &early_res[i];
107 if (last >= r->start && addr < r->end) {
108 *addrp = addr = r->end;
117 * This function checks if any part of the range <start,end> is mapped
121 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
125 for (i = 0; i < e820.nr_map; i++) {
126 struct e820entry *ei = &e820.map[i];
128 if (type && ei->type != type)
130 if (ei->addr >= end || ei->addr + ei->size <= start)
136 EXPORT_SYMBOL_GPL(e820_any_mapped);
139 * This function checks if the entire range <start,end> is mapped with type.
141 * Note: this function only works correct if the e820 table is sorted and
142 * not-overlapping, which is the case
144 int __init e820_all_mapped(unsigned long start, unsigned long end,
149 for (i = 0; i < e820.nr_map; i++) {
150 struct e820entry *ei = &e820.map[i];
152 if (type && ei->type != type)
154 /* is the region (part) in overlap with the current region ?*/
155 if (ei->addr >= end || ei->addr + ei->size <= start)
158 /* if the region is at the beginning of <start,end> we move
159 * start to the end of the region since it's ok until there
161 if (ei->addr <= start)
162 start = ei->addr + ei->size;
164 * if start is now at or beyond end, we're done, full
174 * Find a free area in a specific range.
176 unsigned long __init find_e820_area(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 addr = ei->addr, last;
185 if (ei->type != E820_RAM)
189 if (addr > ei->addr + ei->size)
191 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
193 last = PAGE_ALIGN(addr) + size;
194 if (last > ei->addr + ei->size)
204 * Find the highest page frame number we have available
206 unsigned long __init e820_end_of_ram(void)
208 unsigned long end_pfn;
210 end_pfn = find_max_pfn_with_active_regions();
212 if (end_pfn > end_pfn_map)
213 end_pfn_map = end_pfn;
214 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
215 end_pfn_map = MAXMEM>>PAGE_SHIFT;
216 if (end_pfn > end_user_pfn)
217 end_pfn = end_user_pfn;
218 if (end_pfn > end_pfn_map)
219 end_pfn = end_pfn_map;
221 printk(KERN_INFO "end_pfn_map = %lu\n", end_pfn_map);
226 * Mark e820 reserved areas as busy for the resource manager.
228 void __init e820_reserve_resources(struct resource *code_resource,
229 struct resource *data_resource, struct resource *bss_resource)
232 for (i = 0; i < e820.nr_map; i++) {
233 struct resource *res;
234 res = alloc_bootmem_low(sizeof(struct resource));
235 switch (e820.map[i].type) {
236 case E820_RAM: res->name = "System RAM"; break;
237 case E820_ACPI: res->name = "ACPI Tables"; break;
238 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
239 default: res->name = "reserved";
241 res->start = e820.map[i].addr;
242 res->end = res->start + e820.map[i].size - 1;
243 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
244 request_resource(&iomem_resource, res);
245 if (e820.map[i].type == E820_RAM) {
247 * We don't know which RAM region contains kernel data,
248 * so we try it repeatedly and let the resource manager
251 request_resource(res, code_resource);
252 request_resource(res, data_resource);
253 request_resource(res, bss_resource);
255 if (crashk_res.start != crashk_res.end)
256 request_resource(res, &crashk_res);
263 * Find the ranges of physical addresses that do not correspond to
264 * e820 RAM areas and mark the corresponding pages as nosave for software
265 * suspend and suspend to RAM.
267 * This function requires the e820 map to be sorted and without any
268 * overlapping entries and assumes the first e820 area to be RAM.
270 void __init e820_mark_nosave_regions(void)
275 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
276 for (i = 1; i < e820.nr_map; i++) {
277 struct e820entry *ei = &e820.map[i];
279 if (paddr < ei->addr)
280 register_nosave_region(PFN_DOWN(paddr),
283 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
284 if (ei->type != E820_RAM)
285 register_nosave_region(PFN_UP(ei->addr),
288 if (paddr >= (end_pfn << PAGE_SHIFT))
294 * Finds an active region in the address range from start_pfn to end_pfn and
295 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
297 static int __init e820_find_active_region(const struct e820entry *ei,
298 unsigned long start_pfn,
299 unsigned long end_pfn,
300 unsigned long *ei_startpfn,
301 unsigned long *ei_endpfn)
303 *ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
304 *ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE) >> PAGE_SHIFT;
306 /* Skip map entries smaller than a page */
307 if (*ei_startpfn >= *ei_endpfn)
310 /* Check if end_pfn_map should be updated */
311 if (ei->type != E820_RAM && *ei_endpfn > end_pfn_map)
312 end_pfn_map = *ei_endpfn;
314 /* Skip if map is outside the node */
315 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
316 *ei_startpfn >= end_pfn)
319 /* Check for overlaps */
320 if (*ei_startpfn < start_pfn)
321 *ei_startpfn = start_pfn;
322 if (*ei_endpfn > end_pfn)
323 *ei_endpfn = end_pfn;
325 /* Obey end_user_pfn to save on memmap */
326 if (*ei_startpfn >= end_user_pfn)
328 if (*ei_endpfn > end_user_pfn)
329 *ei_endpfn = end_user_pfn;
334 /* Walk the e820 map and register active regions within a node */
336 e820_register_active_regions(int nid, unsigned long start_pfn,
337 unsigned long end_pfn)
339 unsigned long ei_startpfn;
340 unsigned long ei_endpfn;
343 for (i = 0; i < e820.nr_map; i++)
344 if (e820_find_active_region(&e820.map[i],
346 &ei_startpfn, &ei_endpfn))
347 add_active_range(nid, ei_startpfn, ei_endpfn);
351 * Add a memory region to the kernel e820 map.
353 void __init add_memory_region(unsigned long start, unsigned long size, int type)
358 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
362 e820.map[x].addr = start;
363 e820.map[x].size = size;
364 e820.map[x].type = type;
369 * Find the hole size (in bytes) in the memory range.
370 * @start: starting address of the memory range to scan
371 * @end: ending address of the memory range to scan
373 unsigned long __init e820_hole_size(unsigned long start, unsigned long end)
375 unsigned long start_pfn = start >> PAGE_SHIFT;
376 unsigned long end_pfn = end >> PAGE_SHIFT;
377 unsigned long ei_startpfn, ei_endpfn, ram = 0;
380 for (i = 0; i < e820.nr_map; i++) {
381 if (e820_find_active_region(&e820.map[i],
383 &ei_startpfn, &ei_endpfn))
384 ram += ei_endpfn - ei_startpfn;
386 return end - start - (ram << PAGE_SHIFT);
389 static void __init e820_print_map(char *who)
393 for (i = 0; i < e820.nr_map; i++) {
394 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
395 (unsigned long long) e820.map[i].addr,
397 (e820.map[i].addr + e820.map[i].size));
398 switch (e820.map[i].type) {
400 printk(KERN_CONT "(usable)\n");
403 printk(KERN_CONT "(reserved)\n");
406 printk(KERN_CONT "(ACPI data)\n");
409 printk(KERN_CONT "(ACPI NVS)\n");
412 printk(KERN_CONT "type %u\n", e820.map[i].type);
419 * Sanitize the BIOS e820 map.
421 * Some e820 responses include overlapping entries. The following
422 * replaces the original e820 map with a new one, removing overlaps.
425 static int __init sanitize_e820_map(struct e820entry *biosmap, char *pnr_map)
427 struct change_member {
428 struct e820entry *pbios; /* pointer to original bios entry */
429 unsigned long long addr; /* address for this change point */
431 static struct change_member change_point_list[2*E820MAX] __initdata;
432 static struct change_member *change_point[2*E820MAX] __initdata;
433 static struct e820entry *overlap_list[E820MAX] __initdata;
434 static struct e820entry new_bios[E820MAX] __initdata;
435 struct change_member *change_tmp;
436 unsigned long current_type, last_type;
437 unsigned long long last_addr;
438 int chgidx, still_changing;
441 int old_nr, new_nr, chg_nr;
445 Visually we're performing the following
446 (1,2,3,4 = memory types)...
448 Sample memory map (w/overlaps):
449 ____22__________________
450 ______________________4_
451 ____1111________________
452 _44_____________________
453 11111111________________
454 ____________________33__
455 ___________44___________
456 __________33333_________
457 ______________22________
458 ___________________2222_
459 _________111111111______
460 _____________________11_
461 _________________4______
463 Sanitized equivalent (no overlap):
464 1_______________________
465 _44_____________________
466 ___1____________________
467 ____22__________________
468 ______11________________
469 _________1______________
470 __________3_____________
471 ___________44___________
472 _____________33_________
473 _______________2________
474 ________________1_______
475 _________________4______
476 ___________________2____
477 ____________________33__
478 ______________________4_
481 /* if there's only one memory region, don't bother */
487 /* bail out if we find any unreasonable addresses in bios map */
488 for (i = 0; i < old_nr; i++)
489 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
492 /* create pointers for initial change-point information (for sorting) */
493 for (i = 0; i < 2 * old_nr; i++)
494 change_point[i] = &change_point_list[i];
496 /* record all known change-points (starting and ending addresses),
497 omitting those that are for empty memory regions */
499 for (i = 0; i < old_nr; i++) {
500 if (biosmap[i].size != 0) {
501 change_point[chgidx]->addr = biosmap[i].addr;
502 change_point[chgidx++]->pbios = &biosmap[i];
503 change_point[chgidx]->addr = biosmap[i].addr +
505 change_point[chgidx++]->pbios = &biosmap[i];
510 /* sort change-point list by memory addresses (low -> high) */
512 while (still_changing) {
514 for (i = 1; i < chg_nr; i++) {
515 unsigned long long curaddr, lastaddr;
516 unsigned long long curpbaddr, lastpbaddr;
518 curaddr = change_point[i]->addr;
519 lastaddr = change_point[i - 1]->addr;
520 curpbaddr = change_point[i]->pbios->addr;
521 lastpbaddr = change_point[i - 1]->pbios->addr;
524 * swap entries, when:
526 * curaddr > lastaddr or
527 * curaddr == lastaddr and curaddr == curpbaddr and
528 * lastaddr != lastpbaddr
530 if (curaddr < lastaddr ||
531 (curaddr == lastaddr && curaddr == curpbaddr &&
532 lastaddr != lastpbaddr)) {
533 change_tmp = change_point[i];
534 change_point[i] = change_point[i-1];
535 change_point[i-1] = change_tmp;
541 /* create a new bios memory map, removing overlaps */
542 overlap_entries = 0; /* number of entries in the overlap table */
543 new_bios_entry = 0; /* index for creating new bios map entries */
544 last_type = 0; /* start with undefined memory type */
545 last_addr = 0; /* start with 0 as last starting address */
547 /* loop through change-points, determining affect on the new bios map */
548 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
549 /* keep track of all overlapping bios entries */
550 if (change_point[chgidx]->addr ==
551 change_point[chgidx]->pbios->addr) {
553 * add map entry to overlap list (> 1 entry
554 * implies an overlap)
556 overlap_list[overlap_entries++] =
557 change_point[chgidx]->pbios;
560 * remove entry from list (order independent,
563 for (i = 0; i < overlap_entries; i++) {
564 if (overlap_list[i] ==
565 change_point[chgidx]->pbios)
567 overlap_list[overlap_entries-1];
572 * if there are overlapping entries, decide which
573 * "type" to use (larger value takes precedence --
574 * 1=usable, 2,3,4,4+=unusable)
577 for (i = 0; i < overlap_entries; i++)
578 if (overlap_list[i]->type > current_type)
579 current_type = overlap_list[i]->type;
581 * continue building up new bios map based on this
584 if (current_type != last_type) {
585 if (last_type != 0) {
586 new_bios[new_bios_entry].size =
587 change_point[chgidx]->addr - last_addr;
589 * move forward only if the new size
592 if (new_bios[new_bios_entry].size != 0)
594 * no more space left for new
597 if (++new_bios_entry >= E820MAX)
600 if (current_type != 0) {
601 new_bios[new_bios_entry].addr =
602 change_point[chgidx]->addr;
603 new_bios[new_bios_entry].type = current_type;
604 last_addr = change_point[chgidx]->addr;
606 last_type = current_type;
609 /* retain count for new bios entries */
610 new_nr = new_bios_entry;
612 /* copy new bios mapping into original location */
613 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
620 * Copy the BIOS e820 map into a safe place.
622 * Sanity-check it while we're at it..
624 * If we're lucky and live on a modern system, the setup code
625 * will have given us a memory map that we can use to properly
626 * set up memory. If we aren't, we'll fake a memory map.
628 static int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
630 /* Only one memory region (or negative)? Ignore it */
635 unsigned long start = biosmap->addr;
636 unsigned long size = biosmap->size;
637 unsigned long end = start + size;
638 unsigned long type = biosmap->type;
640 /* Overflow in 64 bits? Ignore the memory map. */
644 add_memory_region(start, size, type);
645 } while (biosmap++, --nr_map);
649 static void early_panic(char *msg)
655 /* We're not void only for x86 32-bit compat */
656 char * __init machine_specific_memory_setup(void)
658 char *who = "BIOS-e820";
660 * Try to copy the BIOS-supplied E820-map.
662 * Otherwise fake a memory map; one section from 0k->640k,
663 * the next section from 1mb->appropriate_mem_k
665 sanitize_e820_map(boot_params.e820_map, &boot_params.e820_entries);
666 if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0)
667 early_panic("Cannot find a valid memory map");
668 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
671 /* In case someone cares... */
675 static int __init parse_memopt(char *p)
679 end_user_pfn = memparse(p, &p);
680 end_user_pfn >>= PAGE_SHIFT;
683 early_param("mem", parse_memopt);
685 static int userdef __initdata;
687 static int __init parse_memmap_opt(char *p)
690 unsigned long long start_at, mem_size;
692 if (!strcmp(p, "exactmap")) {
693 #ifdef CONFIG_CRASH_DUMP
695 * If we are doing a crash dump, we still need to know
696 * the real mem size before original memory map is
699 e820_register_active_regions(0, 0, -1UL);
700 saved_max_pfn = e820_end_of_ram();
701 remove_all_active_ranges();
710 mem_size = memparse(p, &p);
716 start_at = memparse(p+1, &p);
717 add_memory_region(start_at, mem_size, E820_RAM);
718 } else if (*p == '#') {
719 start_at = memparse(p+1, &p);
720 add_memory_region(start_at, mem_size, E820_ACPI);
721 } else if (*p == '$') {
722 start_at = memparse(p+1, &p);
723 add_memory_region(start_at, mem_size, E820_RESERVED);
725 end_user_pfn = (mem_size >> PAGE_SHIFT);
727 return *p == '\0' ? 0 : -EINVAL;
729 early_param("memmap", parse_memmap_opt);
731 void __init finish_e820_parsing(void)
734 char nr = e820.nr_map;
736 if (sanitize_e820_map(e820.map, &nr) < 0)
737 early_panic("Invalid user supplied memory map");
740 printk(KERN_INFO "user-defined physical RAM map:\n");
741 e820_print_map("user");
745 void __init update_e820(void)
749 nr_map = e820.nr_map;
750 if (sanitize_e820_map(e820.map, &nr_map))
752 e820.nr_map = nr_map;
753 printk(KERN_INFO "modified physical RAM map:\n");
754 e820_print_map("modified");
757 unsigned long pci_mem_start = 0xaeedbabe;
758 EXPORT_SYMBOL(pci_mem_start);
761 * Search for the biggest gap in the low 32 bits of the e820
762 * memory space. We pass this space to PCI to assign MMIO resources
763 * for hotplug or unconfigured devices in.
764 * Hopefully the BIOS let enough space left.
766 __init void e820_setup_gap(void)
768 unsigned long gapstart, gapsize, round;
773 last = 0x100000000ull;
774 gapstart = 0x10000000;
778 unsigned long long start = e820.map[i].addr;
779 unsigned long long end = start + e820.map[i].size;
782 * Since "last" is at most 4GB, we know we'll
783 * fit in 32 bits if this condition is true
786 unsigned long gap = last - end;
799 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
800 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
802 KERN_ERR "PCI: Unassigned devices with 32bit resource "
803 "registers may break!\n");
807 * See how much we want to round up: start off with
808 * rounding to the next 1MB area.
811 while ((gapsize >> 4) > round)
813 /* Fun with two's complement */
814 pci_mem_start = (gapstart + round) & -round;
817 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
818 pci_mem_start, gapstart, gapsize);
821 int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
825 if (slot < 0 || slot >= e820.nr_map)
827 for (i = slot; i < e820.nr_map; i++) {
828 if (e820.map[i].type != E820_RAM)
832 if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
834 *addr = e820.map[i].addr;
835 *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
836 max_pfn << PAGE_SHIFT) - *addr;