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>
21 #include <asm/pgtable.h>
24 #include <asm/proto.h>
25 #include <asm/bootsetup.h>
26 #include <asm/sections.h>
28 struct e820map e820 __initdata;
31 * PFN of last memory page.
33 unsigned long end_pfn;
34 EXPORT_SYMBOL(end_pfn);
37 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
38 * The direct mapping extends to end_pfn_map, so that we can directly access
39 * apertures, ACPI and other tables without having to play with fixmaps.
41 unsigned long end_pfn_map;
44 * Last pfn which the user wants to use.
46 static unsigned long __initdata end_user_pfn = MAXMEM>>PAGE_SHIFT;
48 extern struct resource code_resource, data_resource;
50 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
51 static inline int bad_addr(unsigned long *addrp, unsigned long size)
53 unsigned long addr = *addrp, last = addr + size;
55 /* various gunk below that needed for SMP startup */
57 *addrp = PAGE_ALIGN(0x8000);
61 /* direct mapping tables of the kernel */
62 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
63 *addrp = PAGE_ALIGN(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 = PAGE_ALIGN(INITRD_START + INITRD_SIZE);
76 if (last >= __pa_symbol(&_text) && addr < __pa_symbol(&_end)) {
77 *addrp = PAGE_ALIGN(__pa_symbol(&_end));
81 if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
82 *addrp = PAGE_ALIGN(ebda_addr + ebda_size);
87 /* NUMA memory to node map */
88 if (last >= nodemap_addr && addr < nodemap_addr + nodemap_size) {
89 *addrp = nodemap_addr + nodemap_size;
93 /* XXX ramdisk image here? */
98 * This function checks if any part of the range <start,end> is mapped
102 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
105 for (i = 0; i < e820.nr_map; i++) {
106 struct e820entry *ei = &e820.map[i];
107 if (type && ei->type != type)
109 if (ei->addr >= end || ei->addr + ei->size <= start)
117 * This function checks if the entire range <start,end> is mapped with type.
119 * Note: this function only works correct if the e820 table is sorted and
120 * not-overlapping, which is the case
122 int __init e820_all_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];
127 if (type && ei->type != type)
129 /* is the region (part) in overlap with the current region ?*/
130 if (ei->addr >= end || ei->addr + ei->size <= start)
133 /* if the region is at the beginning of <start,end> we move
134 * start to the end of the region since it's ok until there
136 if (ei->addr <= start)
137 start = ei->addr + ei->size;
138 /* if start is now at or beyond end, we're done, full coverage */
140 return 1; /* we're done */
146 * Find a free area in a specific range.
148 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
151 for (i = 0; i < e820.nr_map; i++) {
152 struct e820entry *ei = &e820.map[i];
153 unsigned long addr = ei->addr, last;
154 if (ei->type != E820_RAM)
158 if (addr > ei->addr + ei->size)
160 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
162 last = PAGE_ALIGN(addr) + size;
163 if (last > ei->addr + ei->size)
173 * Find the highest page frame number we have available
175 unsigned long __init e820_end_of_ram(void)
177 unsigned long end_pfn = 0;
178 end_pfn = find_max_pfn_with_active_regions();
180 if (end_pfn > end_pfn_map)
181 end_pfn_map = end_pfn;
182 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
183 end_pfn_map = MAXMEM>>PAGE_SHIFT;
184 if (end_pfn > end_user_pfn)
185 end_pfn = end_user_pfn;
186 if (end_pfn > end_pfn_map)
187 end_pfn = end_pfn_map;
189 printk("end_pfn_map = %lu\n", end_pfn_map);
194 * Mark e820 reserved areas as busy for the resource manager.
196 void __init e820_reserve_resources(void)
199 for (i = 0; i < e820.nr_map; i++) {
200 struct resource *res;
201 res = alloc_bootmem_low(sizeof(struct resource));
202 switch (e820.map[i].type) {
203 case E820_RAM: res->name = "System RAM"; break;
204 case E820_ACPI: res->name = "ACPI Tables"; break;
205 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
206 default: res->name = "reserved";
208 res->start = e820.map[i].addr;
209 res->end = res->start + e820.map[i].size - 1;
210 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
211 request_resource(&iomem_resource, res);
212 if (e820.map[i].type == E820_RAM) {
214 * We don't know which RAM region contains kernel data,
215 * so we try it repeatedly and let the resource manager
218 request_resource(res, &code_resource);
219 request_resource(res, &data_resource);
221 request_resource(res, &crashk_res);
227 /* Mark pages corresponding to given address range as nosave */
229 e820_mark_nosave_range(unsigned long start, unsigned long end)
231 unsigned long pfn, max_pfn;
236 printk("Nosave address range: %016lx - %016lx\n", start, end);
237 max_pfn = end >> PAGE_SHIFT;
238 for (pfn = start >> PAGE_SHIFT; pfn < max_pfn; pfn++)
240 SetPageNosave(pfn_to_page(pfn));
244 * Find the ranges of physical addresses that do not correspond to
245 * e820 RAM areas and mark the corresponding pages as nosave for software
246 * suspend and suspend to RAM.
248 * This function requires the e820 map to be sorted and without any
249 * overlapping entries and assumes the first e820 area to be RAM.
251 void __init e820_mark_nosave_regions(void)
256 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
257 for (i = 1; i < e820.nr_map; i++) {
258 struct e820entry *ei = &e820.map[i];
260 if (paddr < ei->addr)
261 e820_mark_nosave_range(paddr,
262 round_up(ei->addr, PAGE_SIZE));
264 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
265 if (ei->type != E820_RAM)
266 e820_mark_nosave_range(round_up(ei->addr, PAGE_SIZE),
269 if (paddr >= (end_pfn << PAGE_SHIFT))
274 /* Walk the e820 map and register active regions within a node */
276 e820_register_active_regions(int nid, unsigned long start_pfn,
277 unsigned long end_pfn)
280 unsigned long ei_startpfn, ei_endpfn;
281 for (i = 0; i < e820.nr_map; i++) {
282 struct e820entry *ei = &e820.map[i];
283 ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
284 ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE)
287 /* Skip map entries smaller than a page */
288 if (ei_startpfn >= ei_endpfn)
291 /* Check if end_pfn_map should be updated */
292 if (ei->type != E820_RAM && ei_endpfn > end_pfn_map)
293 end_pfn_map = ei_endpfn;
295 /* Skip if map is outside the node */
296 if (ei->type != E820_RAM ||
297 ei_endpfn <= start_pfn ||
298 ei_startpfn >= end_pfn)
301 /* Check for overlaps */
302 if (ei_startpfn < start_pfn)
303 ei_startpfn = start_pfn;
304 if (ei_endpfn > end_pfn)
307 /* Obey end_user_pfn to save on memmap */
308 if (ei_startpfn >= end_user_pfn)
310 if (ei_endpfn > end_user_pfn)
311 ei_endpfn = end_user_pfn;
313 add_active_range(nid, ei_startpfn, ei_endpfn);
318 * Add a memory region to the kernel e820 map.
320 void __init add_memory_region(unsigned long start, unsigned long size, int type)
325 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
329 e820.map[x].addr = start;
330 e820.map[x].size = size;
331 e820.map[x].type = type;
335 void __init e820_print_map(char *who)
339 for (i = 0; i < e820.nr_map; i++) {
340 printk(" %s: %016Lx - %016Lx ", who,
341 (unsigned long long) e820.map[i].addr,
342 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
343 switch (e820.map[i].type) {
344 case E820_RAM: printk("(usable)\n");
347 printk("(reserved)\n");
350 printk("(ACPI data)\n");
353 printk("(ACPI NVS)\n");
355 default: printk("type %u\n", e820.map[i].type);
362 * Sanitize the BIOS e820 map.
364 * Some e820 responses include overlapping entries. The following
365 * replaces the original e820 map with a new one, removing overlaps.
368 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
370 struct change_member {
371 struct e820entry *pbios; /* pointer to original bios entry */
372 unsigned long long addr; /* address for this change point */
374 static struct change_member change_point_list[2*E820MAX] __initdata;
375 static struct change_member *change_point[2*E820MAX] __initdata;
376 static struct e820entry *overlap_list[E820MAX] __initdata;
377 static struct e820entry new_bios[E820MAX] __initdata;
378 struct change_member *change_tmp;
379 unsigned long current_type, last_type;
380 unsigned long long last_addr;
381 int chgidx, still_changing;
384 int old_nr, new_nr, chg_nr;
388 Visually we're performing the following (1,2,3,4 = memory types)...
390 Sample memory map (w/overlaps):
391 ____22__________________
392 ______________________4_
393 ____1111________________
394 _44_____________________
395 11111111________________
396 ____________________33__
397 ___________44___________
398 __________33333_________
399 ______________22________
400 ___________________2222_
401 _________111111111______
402 _____________________11_
403 _________________4______
405 Sanitized equivalent (no overlap):
406 1_______________________
407 _44_____________________
408 ___1____________________
409 ____22__________________
410 ______11________________
411 _________1______________
412 __________3_____________
413 ___________44___________
414 _____________33_________
415 _______________2________
416 ________________1_______
417 _________________4______
418 ___________________2____
419 ____________________33__
420 ______________________4_
423 /* if there's only one memory region, don't bother */
429 /* bail out if we find any unreasonable addresses in bios map */
430 for (i=0; i<old_nr; i++)
431 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
434 /* create pointers for initial change-point information (for sorting) */
435 for (i=0; i < 2*old_nr; i++)
436 change_point[i] = &change_point_list[i];
438 /* record all known change-points (starting and ending addresses),
439 omitting those that are for empty memory regions */
441 for (i=0; i < old_nr; i++) {
442 if (biosmap[i].size != 0) {
443 change_point[chgidx]->addr = biosmap[i].addr;
444 change_point[chgidx++]->pbios = &biosmap[i];
445 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
446 change_point[chgidx++]->pbios = &biosmap[i];
451 /* sort change-point list by memory addresses (low -> high) */
453 while (still_changing) {
455 for (i=1; i < chg_nr; i++) {
456 /* if <current_addr> > <last_addr>, swap */
457 /* or, if current=<start_addr> & last=<end_addr>, swap */
458 if ((change_point[i]->addr < change_point[i-1]->addr) ||
459 ((change_point[i]->addr == change_point[i-1]->addr) &&
460 (change_point[i]->addr == change_point[i]->pbios->addr) &&
461 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
464 change_tmp = change_point[i];
465 change_point[i] = change_point[i-1];
466 change_point[i-1] = change_tmp;
472 /* create a new bios memory map, removing overlaps */
473 overlap_entries=0; /* number of entries in the overlap table */
474 new_bios_entry=0; /* index for creating new bios map entries */
475 last_type = 0; /* start with undefined memory type */
476 last_addr = 0; /* start with 0 as last starting address */
477 /* loop through change-points, determining affect on the new bios map */
478 for (chgidx=0; chgidx < chg_nr; chgidx++)
480 /* keep track of all overlapping bios entries */
481 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
483 /* add map entry to overlap list (> 1 entry implies an overlap) */
484 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
488 /* remove entry from list (order independent, so swap with last) */
489 for (i=0; i<overlap_entries; i++)
491 if (overlap_list[i] == change_point[chgidx]->pbios)
492 overlap_list[i] = overlap_list[overlap_entries-1];
496 /* if there are overlapping entries, decide which "type" to use */
497 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
499 for (i=0; i<overlap_entries; i++)
500 if (overlap_list[i]->type > current_type)
501 current_type = overlap_list[i]->type;
502 /* continue building up new bios map based on this information */
503 if (current_type != last_type) {
504 if (last_type != 0) {
505 new_bios[new_bios_entry].size =
506 change_point[chgidx]->addr - last_addr;
507 /* move forward only if the new size was non-zero */
508 if (new_bios[new_bios_entry].size != 0)
509 if (++new_bios_entry >= E820MAX)
510 break; /* no more space left for new bios entries */
512 if (current_type != 0) {
513 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
514 new_bios[new_bios_entry].type = current_type;
515 last_addr=change_point[chgidx]->addr;
517 last_type = current_type;
520 new_nr = new_bios_entry; /* retain count for new bios entries */
522 /* copy new bios mapping into original location */
523 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
530 * Copy the BIOS e820 map into a safe place.
532 * Sanity-check it while we're at it..
534 * If we're lucky and live on a modern system, the setup code
535 * will have given us a memory map that we can use to properly
536 * set up memory. If we aren't, we'll fake a memory map.
538 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
540 /* Only one memory region (or negative)? Ignore it */
545 unsigned long start = biosmap->addr;
546 unsigned long size = biosmap->size;
547 unsigned long end = start + size;
548 unsigned long type = biosmap->type;
550 /* Overflow in 64 bits? Ignore the memory map. */
554 add_memory_region(start, size, type);
555 } while (biosmap++,--nr_map);
559 void early_panic(char *msg)
565 void __init setup_memory_region(void)
568 * Try to copy the BIOS-supplied E820-map.
570 * Otherwise fake a memory map; one section from 0k->640k,
571 * the next section from 1mb->appropriate_mem_k
573 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
574 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0)
575 early_panic("Cannot find a valid memory map");
576 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
577 e820_print_map("BIOS-e820");
580 static int __init parse_memopt(char *p)
584 end_user_pfn = memparse(p, &p);
585 end_user_pfn >>= PAGE_SHIFT;
588 early_param("mem", parse_memopt);
590 static int userdef __initdata;
592 static int __init parse_memmap_opt(char *p)
595 unsigned long long start_at, mem_size;
597 if (!strcmp(p, "exactmap")) {
598 #ifdef CONFIG_CRASH_DUMP
599 /* If we are doing a crash dump, we
600 * still need to know the real mem
601 * size before original memory map is
604 e820_register_active_regions(0, 0, -1UL);
605 saved_max_pfn = e820_end_of_ram();
606 remove_all_active_ranges();
615 mem_size = memparse(p, &p);
619 start_at = memparse(p+1, &p);
620 add_memory_region(start_at, mem_size, E820_RAM);
621 } else if (*p == '#') {
622 start_at = memparse(p+1, &p);
623 add_memory_region(start_at, mem_size, E820_ACPI);
624 } else if (*p == '$') {
625 start_at = memparse(p+1, &p);
626 add_memory_region(start_at, mem_size, E820_RESERVED);
628 end_user_pfn = (mem_size >> PAGE_SHIFT);
630 return *p == '\0' ? 0 : -EINVAL;
632 early_param("memmap", parse_memmap_opt);
634 void finish_e820_parsing(void)
637 printk(KERN_INFO "user-defined physical RAM map:\n");
638 e820_print_map("user");
642 unsigned long pci_mem_start = 0xaeedbabe;
643 EXPORT_SYMBOL(pci_mem_start);
646 * Search for the biggest gap in the low 32 bits of the e820
647 * memory space. We pass this space to PCI to assign MMIO resources
648 * for hotplug or unconfigured devices in.
649 * Hopefully the BIOS let enough space left.
651 __init void e820_setup_gap(void)
653 unsigned long gapstart, gapsize, round;
658 last = 0x100000000ull;
659 gapstart = 0x10000000;
663 unsigned long long start = e820.map[i].addr;
664 unsigned long long end = start + e820.map[i].size;
667 * Since "last" is at most 4GB, we know we'll
668 * fit in 32 bits if this condition is true
671 unsigned long gap = last - end;
684 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
685 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
686 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
690 * See how much we want to round up: start off with
691 * rounding to the next 1MB area.
694 while ((gapsize >> 4) > round)
696 /* Fun with two's complement */
697 pci_mem_start = (gapstart + round) & -round;
699 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
700 pci_mem_start, gapstart, gapsize);