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>
27 #include <xen/interface/memory.h>
29 struct e820map e820 __initdata;
32 * PFN of last memory page.
34 unsigned long end_pfn;
35 EXPORT_SYMBOL(end_pfn);
38 * end_pfn only includes RAM, while end_pfn_map includes all e820 entries.
39 * The direct mapping extends to end_pfn_map, so that we can directly access
40 * apertures, ACPI and other tables without having to play with fixmaps.
42 unsigned long end_pfn_map;
45 * Last pfn which the user wants to use.
47 static unsigned long __initdata end_user_pfn = MAXMEM>>PAGE_SHIFT;
49 extern struct resource code_resource, data_resource;
51 /* Check for some hardcoded bad areas that early boot is not allowed to touch */
52 static inline int bad_addr(unsigned long *addrp, unsigned long size)
54 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);
88 /* XXX ramdisk image here? */
90 if (last < (table_end<<PAGE_SHIFT)) {
91 *addrp = table_end << PAGE_SHIFT;
99 * This function checks if any part of the range <start,end> is mapped
103 e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
106 for (i = 0; i < e820.nr_map; i++) {
107 struct e820entry *ei = &e820.map[i];
108 if (type && ei->type != type)
110 if (ei->addr >= end || ei->addr + ei->size <= start)
118 * This function checks if the entire range <start,end> is mapped with type.
120 * Note: this function only works correct if the e820 table is sorted and
121 * not-overlapping, which is the case
123 int __init e820_all_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];
128 if (type && ei->type != type)
130 /* is the region (part) in overlap with the current region ?*/
131 if (ei->addr >= end || ei->addr + ei->size <= start)
134 /* if the region is at the beginning of <start,end> we move
135 * start to the end of the region since it's ok until there
137 if (ei->addr <= start)
138 start = ei->addr + ei->size;
139 /* if start is now at or beyond end, we're done, full coverage */
141 return 1; /* we're done */
147 * Find a free area in a specific range.
149 unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size)
152 for (i = 0; i < e820.nr_map; i++) {
153 struct e820entry *ei = &e820.map[i];
154 unsigned long addr = ei->addr, last;
155 if (ei->type != E820_RAM)
159 if (addr > ei->addr + ei->size)
161 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
163 last = PAGE_ALIGN(addr) + size;
164 if (last > ei->addr + ei->size)
174 * Find the highest page frame number we have available
176 unsigned long __init e820_end_of_ram(void)
178 unsigned long end_pfn = 0;
179 end_pfn = find_max_pfn_with_active_regions();
181 if (end_pfn > end_pfn_map)
182 end_pfn_map = end_pfn;
183 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
184 end_pfn_map = MAXMEM>>PAGE_SHIFT;
185 if (end_pfn > end_user_pfn)
186 end_pfn = end_user_pfn;
187 if (end_pfn > end_pfn_map)
188 end_pfn = end_pfn_map;
190 printk("end_pfn_map = %lu\n", end_pfn_map);
195 * Mark e820 reserved areas as busy for the resource manager.
197 void __init e820_reserve_resources(struct e820entry *e820, int nr_map)
200 for (i = 0; i < nr_map; i++) {
201 struct resource *res;
202 res = alloc_bootmem_low(sizeof(struct resource));
203 switch (e820[i].type) {
204 case E820_RAM: res->name = "System RAM"; break;
205 case E820_ACPI: res->name = "ACPI Tables"; break;
206 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
207 default: res->name = "reserved";
209 res->start = e820[i].addr;
210 res->end = res->start + e820[i].size - 1;
211 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
212 request_resource(&iomem_resource, res);
213 if (e820[i].type == E820_RAM) {
215 * We don't know which RAM region contains kernel data,
216 * so we try it repeatedly and let the resource manager
220 request_resource(res, &code_resource);
221 request_resource(res, &data_resource);
224 request_resource(res, &crashk_res);
230 /* Mark pages corresponding to given address range as nosave */
232 e820_mark_nosave_range(unsigned long start, unsigned long end)
234 unsigned long pfn, max_pfn;
239 printk("Nosave address range: %016lx - %016lx\n", start, end);
240 max_pfn = end >> PAGE_SHIFT;
241 for (pfn = start >> PAGE_SHIFT; pfn < max_pfn; pfn++)
243 SetPageNosave(pfn_to_page(pfn));
247 * Find the ranges of physical addresses that do not correspond to
248 * e820 RAM areas and mark the corresponding pages as nosave for software
249 * suspend and suspend to RAM.
251 * This function requires the e820 map to be sorted and without any
252 * overlapping entries and assumes the first e820 area to be RAM.
254 void __init e820_mark_nosave_regions(void)
259 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
260 for (i = 1; i < e820.nr_map; i++) {
261 struct e820entry *ei = &e820.map[i];
263 if (paddr < ei->addr)
264 e820_mark_nosave_range(paddr,
265 round_up(ei->addr, PAGE_SIZE));
267 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
268 if (ei->type != E820_RAM)
269 e820_mark_nosave_range(round_up(ei->addr, PAGE_SIZE),
272 if (paddr >= (end_pfn << PAGE_SHIFT))
277 /* Walk the e820 map and register active regions within a node */
279 e820_register_active_regions(int nid, unsigned long start_pfn,
280 unsigned long end_pfn)
283 unsigned long ei_startpfn, ei_endpfn;
284 for (i = 0; i < e820.nr_map; i++) {
285 struct e820entry *ei = &e820.map[i];
286 ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
287 ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE)
290 /* Skip map entries smaller than a page */
291 if (ei_startpfn >= ei_endpfn)
294 /* Check if end_pfn_map should be updated */
295 if (ei->type != E820_RAM && ei_endpfn > end_pfn_map)
296 end_pfn_map = ei_endpfn;
298 /* Skip if map is outside the node */
299 if (ei->type != E820_RAM ||
300 ei_endpfn <= start_pfn ||
301 ei_startpfn >= end_pfn)
304 /* Check for overlaps */
305 if (ei_startpfn < start_pfn)
306 ei_startpfn = start_pfn;
307 if (ei_endpfn > end_pfn)
310 /* Obey end_user_pfn to save on memmap */
311 if (ei_startpfn >= end_user_pfn)
313 if (ei_endpfn > end_user_pfn)
314 ei_endpfn = end_user_pfn;
316 add_active_range(nid, ei_startpfn, ei_endpfn);
321 * Add a memory region to the kernel e820 map.
323 void __init add_memory_region(unsigned long start, unsigned long size, int type)
328 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
332 e820.map[x].addr = start;
333 e820.map[x].size = size;
334 e820.map[x].type = type;
338 void __init e820_print_map(char *who)
342 for (i = 0; i < e820.nr_map; i++) {
343 printk(" %s: %016Lx - %016Lx ", who,
344 (unsigned long long) e820.map[i].addr,
345 (unsigned long long) (e820.map[i].addr + e820.map[i].size));
346 switch (e820.map[i].type) {
347 case E820_RAM: printk("(usable)\n");
350 printk("(reserved)\n");
353 printk("(ACPI data)\n");
356 printk("(ACPI NVS)\n");
358 default: printk("type %u\n", e820.map[i].type);
365 * Sanitize the BIOS e820 map.
367 * Some e820 responses include overlapping entries. The following
368 * replaces the original e820 map with a new one, removing overlaps.
371 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
373 struct change_member {
374 struct e820entry *pbios; /* pointer to original bios entry */
375 unsigned long long addr; /* address for this change point */
377 static struct change_member change_point_list[2*E820MAX] __initdata;
378 static struct change_member *change_point[2*E820MAX] __initdata;
379 static struct e820entry *overlap_list[E820MAX] __initdata;
380 static struct e820entry new_bios[E820MAX] __initdata;
381 struct change_member *change_tmp;
382 unsigned long current_type, last_type;
383 unsigned long long last_addr;
384 int chgidx, still_changing;
387 int old_nr, new_nr, chg_nr;
391 Visually we're performing the following (1,2,3,4 = memory types)...
393 Sample memory map (w/overlaps):
394 ____22__________________
395 ______________________4_
396 ____1111________________
397 _44_____________________
398 11111111________________
399 ____________________33__
400 ___________44___________
401 __________33333_________
402 ______________22________
403 ___________________2222_
404 _________111111111______
405 _____________________11_
406 _________________4______
408 Sanitized equivalent (no overlap):
409 1_______________________
410 _44_____________________
411 ___1____________________
412 ____22__________________
413 ______11________________
414 _________1______________
415 __________3_____________
416 ___________44___________
417 _____________33_________
418 _______________2________
419 ________________1_______
420 _________________4______
421 ___________________2____
422 ____________________33__
423 ______________________4_
426 /* if there's only one memory region, don't bother */
432 /* bail out if we find any unreasonable addresses in bios map */
433 for (i=0; i<old_nr; i++)
434 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
437 /* create pointers for initial change-point information (for sorting) */
438 for (i=0; i < 2*old_nr; i++)
439 change_point[i] = &change_point_list[i];
441 /* record all known change-points (starting and ending addresses),
442 omitting those that are for empty memory regions */
444 for (i=0; i < old_nr; i++) {
445 if (biosmap[i].size != 0) {
446 change_point[chgidx]->addr = biosmap[i].addr;
447 change_point[chgidx++]->pbios = &biosmap[i];
448 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
449 change_point[chgidx++]->pbios = &biosmap[i];
454 /* sort change-point list by memory addresses (low -> high) */
456 while (still_changing) {
458 for (i=1; i < chg_nr; i++) {
459 /* if <current_addr> > <last_addr>, swap */
460 /* or, if current=<start_addr> & last=<end_addr>, swap */
461 if ((change_point[i]->addr < change_point[i-1]->addr) ||
462 ((change_point[i]->addr == change_point[i-1]->addr) &&
463 (change_point[i]->addr == change_point[i]->pbios->addr) &&
464 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
467 change_tmp = change_point[i];
468 change_point[i] = change_point[i-1];
469 change_point[i-1] = change_tmp;
475 /* create a new bios memory map, removing overlaps */
476 overlap_entries=0; /* number of entries in the overlap table */
477 new_bios_entry=0; /* index for creating new bios map entries */
478 last_type = 0; /* start with undefined memory type */
479 last_addr = 0; /* start with 0 as last starting address */
480 /* loop through change-points, determining affect on the new bios map */
481 for (chgidx=0; chgidx < chg_nr; chgidx++)
483 /* keep track of all overlapping bios entries */
484 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
486 /* add map entry to overlap list (> 1 entry implies an overlap) */
487 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
491 /* remove entry from list (order independent, so swap with last) */
492 for (i=0; i<overlap_entries; i++)
494 if (overlap_list[i] == change_point[chgidx]->pbios)
495 overlap_list[i] = overlap_list[overlap_entries-1];
499 /* if there are overlapping entries, decide which "type" to use */
500 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
502 for (i=0; i<overlap_entries; i++)
503 if (overlap_list[i]->type > current_type)
504 current_type = overlap_list[i]->type;
505 /* continue building up new bios map based on this information */
506 if (current_type != last_type) {
507 if (last_type != 0) {
508 new_bios[new_bios_entry].size =
509 change_point[chgidx]->addr - last_addr;
510 /* move forward only if the new size was non-zero */
511 if (new_bios[new_bios_entry].size != 0)
512 if (++new_bios_entry >= E820MAX)
513 break; /* no more space left for new bios entries */
515 if (current_type != 0) {
516 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
517 new_bios[new_bios_entry].type = current_type;
518 last_addr=change_point[chgidx]->addr;
520 last_type = current_type;
523 new_nr = new_bios_entry; /* retain count for new bios entries */
525 /* copy new bios mapping into original location */
526 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
533 * Copy the BIOS e820 map into a safe place.
535 * Sanity-check it while we're at it..
537 * If we're lucky and live on a modern system, the setup code
538 * will have given us a memory map that we can use to properly
539 * set up memory. If we aren't, we'll fake a memory map.
541 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
544 /* Only one memory region (or negative)? Ignore it */
552 unsigned long start = biosmap->addr;
553 unsigned long size = biosmap->size;
554 unsigned long end = start + size;
555 unsigned long type = biosmap->type;
557 /* Overflow in 64 bits? Ignore the memory map. */
561 add_memory_region(start, size, type);
562 } while (biosmap++,--nr_map);
566 void early_panic(char *msg)
572 void __init setup_memory_region(void)
576 * Try to copy the BIOS-supplied E820-map.
578 * Otherwise fake a memory map; one section from 0k->640k,
579 * the next section from 1mb->appropriate_mem_k
581 sanitize_e820_map(E820_MAP, &E820_MAP_NR);
582 if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0)
583 early_panic("Cannot find a valid memory map");
584 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
585 e820_print_map("BIOS-e820");
586 #else /* CONFIG_XEN */
588 struct xen_memory_map memmap;
590 * This is rather large for a stack variable but this early in
591 * the boot process we know we have plenty slack space.
593 struct e820entry map[E820MAX];
595 memmap.nr_entries = E820MAX;
596 set_xen_guest_handle(memmap.buffer, map);
598 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
599 if ( rc == -ENOSYS ) {
600 memmap.nr_entries = 1;
602 map[0].size = xen_start_info->nr_pages << PAGE_SHIFT;
603 /* 8MB slack (to balance backend allocations). */
604 map[0].size += 8 << 20;
605 map[0].type = E820_RAM;
610 sanitize_e820_map(map, (char *)&memmap.nr_entries);
611 if (copy_e820_map(map, (char)memmap.nr_entries) < 0)
612 early_panic("Cannot find a valid memory map");
613 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
614 e820_print_map("Xen");
618 static int __init parse_memopt(char *p)
621 unsigned long current_end;
626 end_user_pfn = memparse(p, &p);
627 end_user_pfn >>= PAGE_SHIFT;
629 end = end_user_pfn<<PAGE_SHIFT;
631 current_end = e820.map[i].addr + e820.map[i].size;
633 if (current_end < end) {
635 * The e820 map ends before our requested size so
636 * extend the final entry to the requested address.
638 if (e820.map[i].type == E820_RAM)
639 e820.map[i].size = end - e820.map[i].addr;
641 add_memory_region(current_end, end - current_end, E820_RAM);
645 early_param("mem", parse_memopt);
647 static int userdef __initdata;
649 static int __init parse_memmap_opt(char *p)
652 unsigned long long start_at, mem_size;
654 if (!strcmp(p, "exactmap")) {
655 #ifdef CONFIG_CRASH_DUMP
656 /* If we are doing a crash dump, we
657 * still need to know the real mem
658 * size before original memory map is
661 e820_register_active_regions(0, 0, -1UL);
662 saved_max_pfn = e820_end_of_ram();
663 remove_all_active_ranges();
672 mem_size = memparse(p, &p);
676 start_at = memparse(p+1, &p);
677 add_memory_region(start_at, mem_size, E820_RAM);
678 } else if (*p == '#') {
679 start_at = memparse(p+1, &p);
680 add_memory_region(start_at, mem_size, E820_ACPI);
681 } else if (*p == '$') {
682 start_at = memparse(p+1, &p);
683 add_memory_region(start_at, mem_size, E820_RESERVED);
685 end_user_pfn = (mem_size >> PAGE_SHIFT);
687 return *p == '\0' ? 0 : -EINVAL;
689 early_param("memmap", parse_memmap_opt);
691 void finish_e820_parsing(void)
694 printk(KERN_INFO "user-defined physical RAM map:\n");
695 e820_print_map("user");
699 unsigned long pci_mem_start = 0xaeedbabe;
700 EXPORT_SYMBOL(pci_mem_start);
703 * Search for the biggest gap in the low 32 bits of the e820
704 * memory space. We pass this space to PCI to assign MMIO resources
705 * for hotplug or unconfigured devices in.
706 * Hopefully the BIOS let enough space left.
708 __init void e820_setup_gap(struct e820entry *e820, int nr_map)
710 unsigned long gapstart, gapsize, round;
715 last = 0x100000000ull;
716 gapstart = 0x10000000;
720 unsigned long long start = e820[i].addr;
721 unsigned long long end = start + e820[i].size;
724 * Since "last" is at most 4GB, we know we'll
725 * fit in 32 bits if this condition is true
728 unsigned long gap = last - end;
741 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
742 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit address range\n"
743 KERN_ERR "PCI: Unassigned devices with 32bit resource registers may break!\n");
747 * See how much we want to round up: start off with
748 * rounding to the next 1MB area.
751 while ((gapsize >> 4) > round)
753 /* Fun with two's complement */
754 pci_mem_start = (gapstart + round) & -round;
756 printk(KERN_INFO "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
757 pci_mem_start, gapstart, gapsize);