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 */
61 /* direct mapping tables of the kernel */
62 if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) {
63 *addrp = 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 = INITRD_START + INITRD_SIZE;
76 if (last >= __pa_symbol(&_text) && last < __pa_symbol(&_end)) {
77 *addrp = __pa_symbol(&_end);
81 if (last >= ebda_addr && addr < ebda_addr + ebda_size) {
82 *addrp = ebda_addr + ebda_size;
86 /* XXX ramdisk image here? */
91 * This function checks if any part of the range <start,end> is mapped
95 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
98 for (i = 0; i < e820.nr_map; i++) {
99 struct e820entry *ei = &e820.map[i];
100 if (type && ei->type != type)
102 if (ei->addr >= end || ei->addr + ei->size <= start)
110 * This function checks if the entire range <start,end> is mapped with type.
112 * Note: this function only works correct if the e820 table is sorted and
113 * not-overlapping, which is the case
115 int __init e820_all_mapped(unsigned long start, unsigned long end, unsigned type)
118 for (i = 0; i < e820.nr_map; i++) {
119 struct e820entry *ei = &e820.map[i];
120 if (type && ei->type != type)
122 /* is the region (part) in overlap with the current region ?*/
123 if (ei->addr >= end || ei->addr + ei->size <= start)
126 /* if the region is at the beginning of <start,end> we move
127 * start to the end of the region since it's ok until there
129 if (ei->addr <= start)
130 start = ei->addr + ei->size;
131 /* if start is now at or beyond end, we're done, full coverage */
133 return 1; /* we're done */
139 * Find a free area in a specific range.
141 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
144 for (i = 0; i < e820.nr_map; i++) {
145 struct e820entry *ei = &e820.map[i];
146 unsigned long addr = ei->addr, last;
147 if (ei->type != E820_RAM)
151 if (addr > ei->addr + ei->size)
153 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
156 if (last > ei->addr + ei->size)
166 * Find the highest page frame number we have available
168 unsigned long __init e820_end_of_ram(void)
170 unsigned long end_pfn = 0;
171 end_pfn = find_max_pfn_with_active_regions();
173 if (end_pfn > end_pfn_map)
174 end_pfn_map = end_pfn;
175 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
176 end_pfn_map = MAXMEM>>PAGE_SHIFT;
177 if (end_pfn > end_user_pfn)
178 end_pfn = end_user_pfn;
179 if (end_pfn > end_pfn_map)
180 end_pfn = end_pfn_map;
182 printk("end_pfn_map = %lu\n", end_pfn_map);
187 * Mark e820 reserved areas as busy for the resource manager.
189 void __init e820_reserve_resources(void)
192 for (i = 0; i < e820.nr_map; i++) {
193 struct resource *res;
194 res = alloc_bootmem_low(sizeof(struct resource));
195 switch (e820.map[i].type) {
196 case E820_RAM: res->name = "System RAM"; break;
197 case E820_ACPI: res->name = "ACPI Tables"; break;
198 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
199 default: res->name = "reserved";
201 res->start = e820.map[i].addr;
202 res->end = res->start + e820.map[i].size - 1;
203 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
204 request_resource(&iomem_resource, res);
205 if (e820.map[i].type == E820_RAM) {
207 * We don't know which RAM region contains kernel data,
208 * so we try it repeatedly and let the resource manager
211 request_resource(res, &code_resource);
212 request_resource(res, &data_resource);
214 request_resource(res, &crashk_res);
220 /* Mark pages corresponding to given address range as nosave */
222 e820_mark_nosave_range(unsigned long start, unsigned long end)
224 unsigned long pfn, max_pfn;
229 printk("Nosave address range: %016lx - %016lx\n", start, end);
230 max_pfn = end >> PAGE_SHIFT;
231 for (pfn = start >> PAGE_SHIFT; pfn < max_pfn; pfn++)
233 SetPageNosave(pfn_to_page(pfn));
237 * Find the ranges of physical addresses that do not correspond to
238 * e820 RAM areas and mark the corresponding pages as nosave for software
239 * suspend and suspend to RAM.
241 * This function requires the e820 map to be sorted and without any
242 * overlapping entries and assumes the first e820 area to be RAM.
244 void __init e820_mark_nosave_regions(void)
249 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
250 for (i = 1; i < e820.nr_map; i++) {
251 struct e820entry *ei = &e820.map[i];
253 if (paddr < ei->addr)
254 e820_mark_nosave_range(paddr,
255 round_up(ei->addr, PAGE_SIZE));
257 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
258 if (ei->type != E820_RAM)
259 e820_mark_nosave_range(round_up(ei->addr, PAGE_SIZE),
262 if (paddr >= (end_pfn << PAGE_SHIFT))
267 /* Walk the e820 map and register active regions within a node */
269 e820_register_active_regions(int nid, unsigned long start_pfn,
270 unsigned long end_pfn)
273 unsigned long ei_startpfn, ei_endpfn;
274 for (i = 0; i < e820.nr_map; i++) {
275 struct e820entry *ei = &e820.map[i];
276 ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
277 ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE)
280 /* Skip map entries smaller than a page */
281 if (ei_startpfn > ei_endpfn)
284 /* Check if end_pfn_map should be updated */
285 if (ei->type != E820_RAM && ei_endpfn > end_pfn_map)
286 end_pfn_map = ei_endpfn;
288 /* Skip if map is outside the node */
289 if (ei->type != E820_RAM ||
290 ei_endpfn <= start_pfn ||
291 ei_startpfn >= end_pfn)
294 /* Check for overlaps */
295 if (ei_startpfn < start_pfn)
296 ei_startpfn = start_pfn;
297 if (ei_endpfn > end_pfn)
300 /* Obey end_user_pfn to save on memmap */
301 if (ei_startpfn >= end_user_pfn)
303 if (ei_endpfn > end_user_pfn)
304 ei_endpfn = end_user_pfn;
306 add_active_range(nid, ei_startpfn, ei_endpfn);
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 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
533 /* Only one memory region (or negative)? Ignore it */
538 unsigned long start = biosmap->addr;
539 unsigned long size = biosmap->size;
540 unsigned long end = start + size;
541 unsigned long type = biosmap->type;
543 /* Overflow in 64 bits? Ignore the memory map. */
547 add_memory_region(start, size, type);
548 } while (biosmap++,--nr_map);
552 void early_panic(char *msg)
558 void __init setup_memory_region(void)
561 * Try to copy the BIOS-supplied E820-map.
563 * Otherwise fake a memory map; one section from 0k->640k,
564 * the next section from 1mb->appropriate_mem_k
566 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
567 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0)
568 early_panic("Cannot find a valid memory map");
569 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
570 e820_print_map("BIOS-e820");
573 static int __init parse_memopt(char *p)
577 end_user_pfn = memparse(p, &p);
578 end_user_pfn >>= PAGE_SHIFT;
581 early_param("mem", parse_memopt);
583 static int userdef __initdata;
585 static int __init parse_memmap_opt(char *p)
588 unsigned long long start_at, mem_size;
590 if (!strcmp(p, "exactmap")) {
591 #ifdef CONFIG_CRASH_DUMP
592 /* If we are doing a crash dump, we
593 * still need to know the real mem
594 * size before original memory map is
597 saved_max_pfn = e820_end_of_ram();
606 mem_size = memparse(p, &p);
610 start_at = memparse(p+1, &p);
611 add_memory_region(start_at, mem_size, E820_RAM);
612 } else if (*p == '#') {
613 start_at = memparse(p+1, &p);
614 add_memory_region(start_at, mem_size, E820_ACPI);
615 } else if (*p == '$') {
616 start_at = memparse(p+1, &p);
617 add_memory_region(start_at, mem_size, E820_RESERVED);
619 end_user_pfn = (mem_size >> PAGE_SHIFT);
621 return *p == '\0' ? 0 : -EINVAL;
623 early_param("memmap", parse_memmap_opt);
625 void finish_e820_parsing(void)
628 printk(KERN_INFO "user-defined physical RAM map:\n");
629 e820_print_map("user");
633 unsigned long pci_mem_start = 0xaeedbabe;
634 EXPORT_SYMBOL(pci_mem_start);
637 * Search for the biggest gap in the low 32 bits of the e820
638 * memory space. We pass this space to PCI to assign MMIO resources
639 * for hotplug or unconfigured devices in.
640 * Hopefully the BIOS let enough space left.
642 __init void e820_setup_gap(void)
644 unsigned long gapstart, gapsize, round;
649 last = 0x100000000ull;
650 gapstart = 0x10000000;
654 unsigned long long start = e820.map[i].addr;
655 unsigned long long end = start + e820.map[i].size;
658 * Since "last" is at most 4GB, we know we'll
659 * fit in 32 bits if this condition is true
662 unsigned long gap = last - end;
675 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
676 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
677 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
681 * See how much we want to round up: start off with
682 * rounding to the next 1MB area.
685 while ((gapsize >> 4) > round)
687 /* Fun with two's complement */
688 pci_mem_start = (gapstart + round) & -round;
690 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
691 pci_mem_start, gapstart, gapsize);