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/bootsetup.h>
28 #include <asm/sections.h>
33 * PFN of last memory page.
35 unsigned long end_pfn;
36 EXPORT_SYMBOL(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;
50 extern struct resource code_resource, data_resource;
52 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
53 static inline int bad_addr(unsigned long *addrp, unsigned long size)
55 unsigned long addr = *addrp, last = addr + size;
57 /* various gunk below that needed for SMP startup */
59 *addrp = PAGE_ALIGN(0x8000);
63 /* direct mapping tables of the kernel */
64 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
65 *addrp = PAGE_ALIGN(table_end << PAGE_SHIFT);
70 #ifdef CONFIG_BLK_DEV_INITRD
71 if (LOADER_TYPE && INITRD_START && last >= INITRD_START &&
72 addr < INITRD_START+INITRD_SIZE) {
73 *addrp = PAGE_ALIGN(INITRD_START + INITRD_SIZE);
78 if (last >= __pa_symbol(&_text) && addr < __pa_symbol(&_end)) {
79 *addrp = PAGE_ALIGN(__pa_symbol(&_end));
83 if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
84 *addrp = PAGE_ALIGN(ebda_addr + ebda_size);
89 /* NUMA memory to node map */
90 if (last >= nodemap_addr && addr < nodemap_addr + nodemap_size) {
91 *addrp = nodemap_addr + nodemap_size;
95 /* XXX ramdisk image here? */
100 * This function checks if any part of the range <start,end> is mapped
104 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
107 for (i = 0; i < e820.nr_map; i++) {
108 struct e820entry *ei = &e820.map[i];
109 if (type && ei->type != type)
111 if (ei->addr >= end || ei->addr + ei->size <= start)
117 EXPORT_SYMBOL_GPL(e820_any_mapped);
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)
165 last = PAGE_ALIGN(addr) + size;
166 if (last > ei->addr + ei->size)
176 * Find the highest page frame number we have available
178 unsigned long __init e820_end_of_ram(void)
180 unsigned long end_pfn = 0;
181 end_pfn = find_max_pfn_with_active_regions();
183 if (end_pfn > end_pfn_map)
184 end_pfn_map = end_pfn;
185 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
186 end_pfn_map = MAXMEM>>PAGE_SHIFT;
187 if (end_pfn > end_user_pfn)
188 end_pfn = end_user_pfn;
189 if (end_pfn > end_pfn_map)
190 end_pfn = end_pfn_map;
192 printk("end_pfn_map = %lu\n", end_pfn_map);
197 * Mark e820 reserved areas as busy for the resource manager.
199 void __init e820_reserve_resources(void)
202 for (i = 0; i < e820.nr_map; i++) {
203 struct resource *res;
204 res = alloc_bootmem_low(sizeof(struct resource));
205 switch (e820.map[i].type) {
206 case E820_RAM: res->name = "System RAM"; break;
207 case E820_ACPI: res->name = "ACPI Tables"; break;
208 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
209 default: res->name = "reserved";
211 res->start = e820.map[i].addr;
212 res->end = res->start + e820.map[i].size - 1;
213 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
214 request_resource(&iomem_resource, res);
215 if (e820.map[i].type == E820_RAM) {
217 * We don't know which RAM region contains kernel data,
218 * so we try it repeatedly and let the resource manager
221 request_resource(res, &code_resource);
222 request_resource(res, &data_resource);
224 request_resource(res, &crashk_res);
231 * Find the ranges of physical addresses that do not correspond to
232 * e820 RAM areas and mark the corresponding pages as nosave for software
233 * suspend and suspend to RAM.
235 * This function requires the e820 map to be sorted and without any
236 * overlapping entries and assumes the first e820 area to be RAM.
238 void __init e820_mark_nosave_regions(void)
243 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
244 for (i = 1; i < e820.nr_map; i++) {
245 struct e820entry *ei = &e820.map[i];
247 if (paddr < ei->addr)
248 register_nosave_region(PFN_DOWN(paddr),
251 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
252 if (ei->type != E820_RAM)
253 register_nosave_region(PFN_UP(ei->addr),
256 if (paddr >= (end_pfn << PAGE_SHIFT))
262 * Finds an active region in the address range from start_pfn to end_pfn and
263 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
265 static int __init e820_find_active_region(const struct e820entry *ei,
266 unsigned long start_pfn,
267 unsigned long end_pfn,
268 unsigned long *ei_startpfn,
269 unsigned long *ei_endpfn)
271 *ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
272 *ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE) >> PAGE_SHIFT;
274 /* Skip map entries smaller than a page */
275 if (*ei_startpfn >= *ei_endpfn)
278 /* Check if end_pfn_map should be updated */
279 if (ei->type != E820_RAM && *ei_endpfn > end_pfn_map)
280 end_pfn_map = *ei_endpfn;
282 /* Skip if map is outside the node */
283 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
284 *ei_startpfn >= end_pfn)
287 /* Check for overlaps */
288 if (*ei_startpfn < start_pfn)
289 *ei_startpfn = start_pfn;
290 if (*ei_endpfn > end_pfn)
291 *ei_endpfn = end_pfn;
293 /* Obey end_user_pfn to save on memmap */
294 if (*ei_startpfn >= end_user_pfn)
296 if (*ei_endpfn > end_user_pfn)
297 *ei_endpfn = end_user_pfn;
302 /* Walk the e820 map and register active regions within a node */
304 e820_register_active_regions(int nid, unsigned long start_pfn,
305 unsigned long end_pfn)
307 unsigned long ei_startpfn;
308 unsigned long ei_endpfn;
311 for (i = 0; i < e820.nr_map; i++)
312 if (e820_find_active_region(&e820.map[i],
314 &ei_startpfn, &ei_endpfn))
315 add_active_range(nid, ei_startpfn, ei_endpfn);
319 * Add a memory region to the kernel e820 map.
321 void __init add_memory_region(unsigned long start, unsigned long size, int type)
326 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
330 e820.map[x].addr = start;
331 e820.map[x].size = size;
332 e820.map[x].type = type;
337 * Find the hole size (in bytes) in the memory range.
338 * @start: starting address of the memory range to scan
339 * @end: ending address of the memory range to scan
341 unsigned long __init e820_hole_size(unsigned long start, unsigned long end)
343 unsigned long start_pfn = start >> PAGE_SHIFT;
344 unsigned long end_pfn = end >> PAGE_SHIFT;
345 unsigned long ei_startpfn;
346 unsigned long ei_endpfn;
347 unsigned long ram = 0;
350 for (i = 0; i < e820.nr_map; i++) {
351 if (e820_find_active_region(&e820.map[i],
353 &ei_startpfn, &ei_endpfn))
354 ram += ei_endpfn - ei_startpfn;
356 return end - start - (ram << PAGE_SHIFT);
359 void __init e820_print_map(char *who)
363 for (i = 0; i < e820.nr_map; i++) {
364 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
365 (unsigned long long) e820.map[i].addr,
366 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
367 switch (e820.map[i].type) {
368 case E820_RAM: printk("(usable)\n");
371 printk("(reserved)\n");
374 printk("(ACPI data)\n");
377 printk("(ACPI NVS)\n");
379 default: printk("type %u\n", e820.map[i].type);
386 * Sanitize the BIOS e820 map.
388 * Some e820 responses include overlapping entries. The following
389 * replaces the original e820 map with a new one, removing overlaps.
392 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
394 struct change_member {
395 struct e820entry *pbios; /* pointer to original bios entry */
396 unsigned long long addr; /* address for this change point */
398 static struct change_member change_point_list[2*E820MAX] __initdata;
399 static struct change_member *change_point[2*E820MAX] __initdata;
400 static struct e820entry *overlap_list[E820MAX] __initdata;
401 static struct e820entry new_bios[E820MAX] __initdata;
402 struct change_member *change_tmp;
403 unsigned long current_type, last_type;
404 unsigned long long last_addr;
405 int chgidx, still_changing;
408 int old_nr, new_nr, chg_nr;
412 Visually we're performing the following (1,2,3,4 = memory types)...
414 Sample memory map (w/overlaps):
415 ____22__________________
416 ______________________4_
417 ____1111________________
418 _44_____________________
419 11111111________________
420 ____________________33__
421 ___________44___________
422 __________33333_________
423 ______________22________
424 ___________________2222_
425 _________111111111______
426 _____________________11_
427 _________________4______
429 Sanitized equivalent (no overlap):
430 1_______________________
431 _44_____________________
432 ___1____________________
433 ____22__________________
434 ______11________________
435 _________1______________
436 __________3_____________
437 ___________44___________
438 _____________33_________
439 _______________2________
440 ________________1_______
441 _________________4______
442 ___________________2____
443 ____________________33__
444 ______________________4_
447 /* if there's only one memory region, don't bother */
453 /* bail out if we find any unreasonable addresses in bios map */
454 for (i=0; i<old_nr; i++)
455 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
458 /* create pointers for initial change-point information (for sorting) */
459 for (i=0; i < 2*old_nr; i++)
460 change_point[i] = &change_point_list[i];
462 /* record all known change-points (starting and ending addresses),
463 omitting those that are for empty memory regions */
465 for (i=0; i < old_nr; i++) {
466 if (biosmap[i].size != 0) {
467 change_point[chgidx]->addr = biosmap[i].addr;
468 change_point[chgidx++]->pbios = &biosmap[i];
469 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
470 change_point[chgidx++]->pbios = &biosmap[i];
475 /* sort change-point list by memory addresses (low -> high) */
477 while (still_changing) {
479 for (i=1; i < chg_nr; i++) {
480 /* if <current_addr> > <last_addr>, swap */
481 /* or, if current=<start_addr> & last=<end_addr>, swap */
482 if ((change_point[i]->addr < change_point[i-1]->addr) ||
483 ((change_point[i]->addr == change_point[i-1]->addr) &&
484 (change_point[i]->addr == change_point[i]->pbios->addr) &&
485 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
488 change_tmp = change_point[i];
489 change_point[i] = change_point[i-1];
490 change_point[i-1] = change_tmp;
496 /* create a new bios memory map, removing overlaps */
497 overlap_entries=0; /* number of entries in the overlap table */
498 new_bios_entry=0; /* index for creating new bios map entries */
499 last_type = 0; /* start with undefined memory type */
500 last_addr = 0; /* start with 0 as last starting address */
501 /* loop through change-points, determining affect on the new bios map */
502 for (chgidx=0; chgidx < chg_nr; chgidx++)
504 /* keep track of all overlapping bios entries */
505 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
507 /* add map entry to overlap list (> 1 entry implies an overlap) */
508 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
512 /* remove entry from list (order independent, so swap with last) */
513 for (i=0; i<overlap_entries; i++)
515 if (overlap_list[i] == change_point[chgidx]->pbios)
516 overlap_list[i] = overlap_list[overlap_entries-1];
520 /* if there are overlapping entries, decide which "type" to use */
521 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
523 for (i=0; i<overlap_entries; i++)
524 if (overlap_list[i]->type > current_type)
525 current_type = overlap_list[i]->type;
526 /* continue building up new bios map based on this information */
527 if (current_type != last_type) {
528 if (last_type != 0) {
529 new_bios[new_bios_entry].size =
530 change_point[chgidx]->addr - last_addr;
531 /* move forward only if the new size was non-zero */
532 if (new_bios[new_bios_entry].size != 0)
533 if (++new_bios_entry >= E820MAX)
534 break; /* no more space left for new bios entries */
536 if (current_type != 0) {
537 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
538 new_bios[new_bios_entry].type = current_type;
539 last_addr=change_point[chgidx]->addr;
541 last_type = current_type;
544 new_nr = new_bios_entry; /* retain count for new bios entries */
546 /* copy new bios mapping into original location */
547 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
554 * Copy the BIOS e820 map into a safe place.
556 * Sanity-check it while we're at it..
558 * If we're lucky and live on a modern system, the setup code
559 * will have given us a memory map that we can use to properly
560 * set up memory. If we aren't, we'll fake a memory map.
562 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
564 /* Only one memory region (or negative)? Ignore it */
569 unsigned long start = biosmap->addr;
570 unsigned long size = biosmap->size;
571 unsigned long end = start + size;
572 unsigned long type = biosmap->type;
574 /* Overflow in 64 bits? Ignore the memory map. */
578 add_memory_region(start, size, type);
579 } while (biosmap++,--nr_map);
583 void early_panic(char *msg)
589 void __init setup_memory_region(void)
592 * Try to copy the BIOS-supplied E820-map.
594 * Otherwise fake a memory map; one section from 0k->640k,
595 * the next section from 1mb->appropriate_mem_k
597 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
598 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0)
599 early_panic("Cannot find a valid memory map");
600 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
601 e820_print_map("BIOS-e820");
604 static int __init parse_memopt(char *p)
608 end_user_pfn = memparse(p, &p);
609 end_user_pfn >>= PAGE_SHIFT;
612 early_param("mem", parse_memopt);
614 static int userdef __initdata;
616 static int __init parse_memmap_opt(char *p)
619 unsigned long long start_at, mem_size;
621 if (!strcmp(p, "exactmap")) {
622 #ifdef CONFIG_CRASH_DUMP
623 /* If we are doing a crash dump, we
624 * still need to know the real mem
625 * size before original memory map is
628 e820_register_active_regions(0, 0, -1UL);
629 saved_max_pfn = e820_end_of_ram();
630 remove_all_active_ranges();
639 mem_size = memparse(p, &p);
643 start_at = memparse(p+1, &p);
644 add_memory_region(start_at, mem_size, E820_RAM);
645 } else if (*p == '#') {
646 start_at = memparse(p+1, &p);
647 add_memory_region(start_at, mem_size, E820_ACPI);
648 } else if (*p == '$') {
649 start_at = memparse(p+1, &p);
650 add_memory_region(start_at, mem_size, E820_RESERVED);
652 end_user_pfn = (mem_size >> PAGE_SHIFT);
654 return *p == '\0' ? 0 : -EINVAL;
656 early_param("memmap", parse_memmap_opt);
658 void __init finish_e820_parsing(void)
661 printk(KERN_INFO "user-defined physical RAM map:\n");
662 e820_print_map("user");
666 unsigned long pci_mem_start = 0xaeedbabe;
667 EXPORT_SYMBOL(pci_mem_start);
670 * Search for the biggest gap in the low 32 bits of the e820
671 * memory space. We pass this space to PCI to assign MMIO resources
672 * for hotplug or unconfigured devices in.
673 * Hopefully the BIOS let enough space left.
675 __init void e820_setup_gap(void)
677 unsigned long gapstart, gapsize, round;
682 last = 0x100000000ull;
683 gapstart = 0x10000000;
687 unsigned long long start = e820.map[i].addr;
688 unsigned long long end = start + e820.map[i].size;
691 * Since "last" is at most 4GB, we know we'll
692 * fit in 32 bits if this condition is true
695 unsigned long gap = last - end;
708 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
709 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
710 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
714 * See how much we want to round up: start off with
715 * rounding to the next 1MB area.
718 while ((gapsize >> 4) > round)
720 /* Fun with two's complement */
721 pci_mem_start = (gapstart + round) & -round;
723 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
724 pci_mem_start, gapstart, gapsize);