iommu/amd: Don't use MSI address range for DMA addresses
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kernel / e820.c
blob28b09af9775c1f73e6dd5e33a0989546abfe048a
1 /*
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/pfn.h>
16 #include <linux/suspend.h>
17 #include <linux/firmware-map.h>
19 #include <asm/e820.h>
20 #include <asm/proto.h>
21 #include <asm/setup.h>
24 * The e820 map is the map that gets modified e.g. with command line parameters
25 * and that is also registered with modifications in the kernel resource tree
26 * with the iomem_resource as parent.
28 * The e820_saved is directly saved after the BIOS-provided memory map is
29 * copied. It doesn't get modified afterwards. It's registered for the
30 * /sys/firmware/memmap interface.
32 * That memory map is not modified and is used as base for kexec. The kexec'd
33 * kernel should get the same memory map as the firmware provides. Then the
34 * user can e.g. boot the original kernel with mem=1G while still booting the
35 * next kernel with full memory.
37 struct e820map e820;
38 struct e820map e820_saved;
40 /* For PCI or other memory-mapped resources */
41 unsigned long pci_mem_start = 0xaeedbabe;
42 #ifdef CONFIG_PCI
43 EXPORT_SYMBOL(pci_mem_start);
44 #endif
47 * This function checks if any part of the range <start,end> is mapped
48 * with type.
50 int
51 e820_any_mapped(u64 start, u64 end, unsigned type)
53 int i;
55 for (i = 0; i < e820.nr_map; i++) {
56 struct e820entry *ei = &e820.map[i];
58 if (type && ei->type != type)
59 continue;
60 if (ei->addr >= end || ei->addr + ei->size <= start)
61 continue;
62 return 1;
64 return 0;
66 EXPORT_SYMBOL_GPL(e820_any_mapped);
69 * This function checks if the entire range <start,end> is mapped with type.
71 * Note: this function only works correct if the e820 table is sorted and
72 * not-overlapping, which is the case
74 int __init e820_all_mapped(u64 start, u64 end, unsigned type)
76 int i;
78 for (i = 0; i < e820.nr_map; i++) {
79 struct e820entry *ei = &e820.map[i];
81 if (type && ei->type != type)
82 continue;
83 /* is the region (part) in overlap with the current region ?*/
84 if (ei->addr >= end || ei->addr + ei->size <= start)
85 continue;
87 /* if the region is at the beginning of <start,end> we move
88 * start to the end of the region since it's ok until there
90 if (ei->addr <= start)
91 start = ei->addr + ei->size;
93 * if start is now at or beyond end, we're done, full
94 * coverage
96 if (start >= end)
97 return 1;
99 return 0;
103 * Add a memory region to the kernel e820 map.
105 static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
106 int type)
108 int x = e820x->nr_map;
110 if (x >= ARRAY_SIZE(e820x->map)) {
111 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
112 return;
115 e820x->map[x].addr = start;
116 e820x->map[x].size = size;
117 e820x->map[x].type = type;
118 e820x->nr_map++;
121 void __init e820_add_region(u64 start, u64 size, int type)
123 __e820_add_region(&e820, start, size, type);
126 static void __init e820_print_type(u32 type)
128 switch (type) {
129 case E820_RAM:
130 case E820_RESERVED_KERN:
131 printk(KERN_CONT "(usable)");
132 break;
133 case E820_RESERVED:
134 printk(KERN_CONT "(reserved)");
135 break;
136 case E820_ACPI:
137 printk(KERN_CONT "(ACPI data)");
138 break;
139 case E820_NVS:
140 printk(KERN_CONT "(ACPI NVS)");
141 break;
142 case E820_UNUSABLE:
143 printk(KERN_CONT "(unusable)");
144 break;
145 default:
146 printk(KERN_CONT "type %u", type);
147 break;
151 void __init e820_print_map(char *who)
153 int i;
155 for (i = 0; i < e820.nr_map; i++) {
156 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
157 (unsigned long long) e820.map[i].addr,
158 (unsigned long long)
159 (e820.map[i].addr + e820.map[i].size));
160 e820_print_type(e820.map[i].type);
161 printk(KERN_CONT "\n");
166 * Sanitize the BIOS e820 map.
168 * Some e820 responses include overlapping entries. The following
169 * replaces the original e820 map with a new one, removing overlaps,
170 * and resolving conflicting memory types in favor of highest
171 * numbered type.
173 * The input parameter biosmap points to an array of 'struct
174 * e820entry' which on entry has elements in the range [0, *pnr_map)
175 * valid, and which has space for up to max_nr_map entries.
176 * On return, the resulting sanitized e820 map entries will be in
177 * overwritten in the same location, starting at biosmap.
179 * The integer pointed to by pnr_map must be valid on entry (the
180 * current number of valid entries located at biosmap) and will
181 * be updated on return, with the new number of valid entries
182 * (something no more than max_nr_map.)
184 * The return value from sanitize_e820_map() is zero if it
185 * successfully 'sanitized' the map entries passed in, and is -1
186 * if it did nothing, which can happen if either of (1) it was
187 * only passed one map entry, or (2) any of the input map entries
188 * were invalid (start + size < start, meaning that the size was
189 * so big the described memory range wrapped around through zero.)
191 * Visually we're performing the following
192 * (1,2,3,4 = memory types)...
194 * Sample memory map (w/overlaps):
195 * ____22__________________
196 * ______________________4_
197 * ____1111________________
198 * _44_____________________
199 * 11111111________________
200 * ____________________33__
201 * ___________44___________
202 * __________33333_________
203 * ______________22________
204 * ___________________2222_
205 * _________111111111______
206 * _____________________11_
207 * _________________4______
209 * Sanitized equivalent (no overlap):
210 * 1_______________________
211 * _44_____________________
212 * ___1____________________
213 * ____22__________________
214 * ______11________________
215 * _________1______________
216 * __________3_____________
217 * ___________44___________
218 * _____________33_________
219 * _______________2________
220 * ________________1_______
221 * _________________4______
222 * ___________________2____
223 * ____________________33__
224 * ______________________4_
227 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
228 u32 *pnr_map)
230 struct change_member {
231 struct e820entry *pbios; /* pointer to original bios entry */
232 unsigned long long addr; /* address for this change point */
234 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
235 static struct change_member *change_point[2*E820_X_MAX] __initdata;
236 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
237 static struct e820entry new_bios[E820_X_MAX] __initdata;
238 struct change_member *change_tmp;
239 unsigned long current_type, last_type;
240 unsigned long long last_addr;
241 int chgidx, still_changing;
242 int overlap_entries;
243 int new_bios_entry;
244 int old_nr, new_nr, chg_nr;
245 int i;
247 /* if there's only one memory region, don't bother */
248 if (*pnr_map < 2)
249 return -1;
251 old_nr = *pnr_map;
252 BUG_ON(old_nr > max_nr_map);
254 /* bail out if we find any unreasonable addresses in bios map */
255 for (i = 0; i < old_nr; i++)
256 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
257 return -1;
259 /* create pointers for initial change-point information (for sorting) */
260 for (i = 0; i < 2 * old_nr; i++)
261 change_point[i] = &change_point_list[i];
263 /* record all known change-points (starting and ending addresses),
264 omitting those that are for empty memory regions */
265 chgidx = 0;
266 for (i = 0; i < old_nr; i++) {
267 if (biosmap[i].size != 0) {
268 change_point[chgidx]->addr = biosmap[i].addr;
269 change_point[chgidx++]->pbios = &biosmap[i];
270 change_point[chgidx]->addr = biosmap[i].addr +
271 biosmap[i].size;
272 change_point[chgidx++]->pbios = &biosmap[i];
275 chg_nr = chgidx;
277 /* sort change-point list by memory addresses (low -> high) */
278 still_changing = 1;
279 while (still_changing) {
280 still_changing = 0;
281 for (i = 1; i < chg_nr; i++) {
282 unsigned long long curaddr, lastaddr;
283 unsigned long long curpbaddr, lastpbaddr;
285 curaddr = change_point[i]->addr;
286 lastaddr = change_point[i - 1]->addr;
287 curpbaddr = change_point[i]->pbios->addr;
288 lastpbaddr = change_point[i - 1]->pbios->addr;
291 * swap entries, when:
293 * curaddr > lastaddr or
294 * curaddr == lastaddr and curaddr == curpbaddr and
295 * lastaddr != lastpbaddr
297 if (curaddr < lastaddr ||
298 (curaddr == lastaddr && curaddr == curpbaddr &&
299 lastaddr != lastpbaddr)) {
300 change_tmp = change_point[i];
301 change_point[i] = change_point[i-1];
302 change_point[i-1] = change_tmp;
303 still_changing = 1;
308 /* create a new bios memory map, removing overlaps */
309 overlap_entries = 0; /* number of entries in the overlap table */
310 new_bios_entry = 0; /* index for creating new bios map entries */
311 last_type = 0; /* start with undefined memory type */
312 last_addr = 0; /* start with 0 as last starting address */
314 /* loop through change-points, determining affect on the new bios map */
315 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
316 /* keep track of all overlapping bios entries */
317 if (change_point[chgidx]->addr ==
318 change_point[chgidx]->pbios->addr) {
320 * add map entry to overlap list (> 1 entry
321 * implies an overlap)
323 overlap_list[overlap_entries++] =
324 change_point[chgidx]->pbios;
325 } else {
327 * remove entry from list (order independent,
328 * so swap with last)
330 for (i = 0; i < overlap_entries; i++) {
331 if (overlap_list[i] ==
332 change_point[chgidx]->pbios)
333 overlap_list[i] =
334 overlap_list[overlap_entries-1];
336 overlap_entries--;
339 * if there are overlapping entries, decide which
340 * "type" to use (larger value takes precedence --
341 * 1=usable, 2,3,4,4+=unusable)
343 current_type = 0;
344 for (i = 0; i < overlap_entries; i++)
345 if (overlap_list[i]->type > current_type)
346 current_type = overlap_list[i]->type;
348 * continue building up new bios map based on this
349 * information
351 if (current_type != last_type) {
352 if (last_type != 0) {
353 new_bios[new_bios_entry].size =
354 change_point[chgidx]->addr - last_addr;
356 * move forward only if the new size
357 * was non-zero
359 if (new_bios[new_bios_entry].size != 0)
361 * no more space left for new
362 * bios entries ?
364 if (++new_bios_entry >= max_nr_map)
365 break;
367 if (current_type != 0) {
368 new_bios[new_bios_entry].addr =
369 change_point[chgidx]->addr;
370 new_bios[new_bios_entry].type = current_type;
371 last_addr = change_point[chgidx]->addr;
373 last_type = current_type;
376 /* retain count for new bios entries */
377 new_nr = new_bios_entry;
379 /* copy new bios mapping into original location */
380 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
381 *pnr_map = new_nr;
383 return 0;
386 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
388 while (nr_map) {
389 u64 start = biosmap->addr;
390 u64 size = biosmap->size;
391 u64 end = start + size;
392 u32 type = biosmap->type;
394 /* Overflow in 64 bits? Ignore the memory map. */
395 if (start > end)
396 return -1;
398 e820_add_region(start, size, type);
400 biosmap++;
401 nr_map--;
403 return 0;
407 * Copy the BIOS e820 map into a safe place.
409 * Sanity-check it while we're at it..
411 * If we're lucky and live on a modern system, the setup code
412 * will have given us a memory map that we can use to properly
413 * set up memory. If we aren't, we'll fake a memory map.
415 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
417 /* Only one memory region (or negative)? Ignore it */
418 if (nr_map < 2)
419 return -1;
421 return __append_e820_map(biosmap, nr_map);
424 static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
425 u64 size, unsigned old_type,
426 unsigned new_type)
428 u64 end;
429 unsigned int i;
430 u64 real_updated_size = 0;
432 BUG_ON(old_type == new_type);
434 if (size > (ULLONG_MAX - start))
435 size = ULLONG_MAX - start;
437 end = start + size;
438 printk(KERN_DEBUG "e820 update range: %016Lx - %016Lx ",
439 (unsigned long long) start,
440 (unsigned long long) end);
441 e820_print_type(old_type);
442 printk(KERN_CONT " ==> ");
443 e820_print_type(new_type);
444 printk(KERN_CONT "\n");
446 for (i = 0; i < e820x->nr_map; i++) {
447 struct e820entry *ei = &e820x->map[i];
448 u64 final_start, final_end;
449 u64 ei_end;
451 if (ei->type != old_type)
452 continue;
454 ei_end = ei->addr + ei->size;
455 /* totally covered by new range? */
456 if (ei->addr >= start && ei_end <= end) {
457 ei->type = new_type;
458 real_updated_size += ei->size;
459 continue;
462 /* new range is totally covered? */
463 if (ei->addr < start && ei_end > end) {
464 __e820_add_region(e820x, start, size, new_type);
465 __e820_add_region(e820x, end, ei_end - end, ei->type);
466 ei->size = start - ei->addr;
467 real_updated_size += size;
468 continue;
471 /* partially covered */
472 final_start = max(start, ei->addr);
473 final_end = min(end, ei_end);
474 if (final_start >= final_end)
475 continue;
477 __e820_add_region(e820x, final_start, final_end - final_start,
478 new_type);
480 real_updated_size += final_end - final_start;
483 * left range could be head or tail, so need to update
484 * size at first.
486 ei->size -= final_end - final_start;
487 if (ei->addr < final_start)
488 continue;
489 ei->addr = final_end;
491 return real_updated_size;
494 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
495 unsigned new_type)
497 return __e820_update_range(&e820, start, size, old_type, new_type);
500 static u64 __init e820_update_range_saved(u64 start, u64 size,
501 unsigned old_type, unsigned new_type)
503 return __e820_update_range(&e820_saved, start, size, old_type,
504 new_type);
507 /* make e820 not cover the range */
508 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
509 int checktype)
511 int i;
512 u64 end;
513 u64 real_removed_size = 0;
515 if (size > (ULLONG_MAX - start))
516 size = ULLONG_MAX - start;
518 end = start + size;
519 printk(KERN_DEBUG "e820 remove range: %016Lx - %016Lx ",
520 (unsigned long long) start,
521 (unsigned long long) end);
522 if (checktype)
523 e820_print_type(old_type);
524 printk(KERN_CONT "\n");
526 for (i = 0; i < e820.nr_map; i++) {
527 struct e820entry *ei = &e820.map[i];
528 u64 final_start, final_end;
529 u64 ei_end;
531 if (checktype && ei->type != old_type)
532 continue;
534 ei_end = ei->addr + ei->size;
535 /* totally covered? */
536 if (ei->addr >= start && ei_end <= end) {
537 real_removed_size += ei->size;
538 memset(ei, 0, sizeof(struct e820entry));
539 continue;
542 /* new range is totally covered? */
543 if (ei->addr < start && ei_end > end) {
544 e820_add_region(end, ei_end - end, ei->type);
545 ei->size = start - ei->addr;
546 real_removed_size += size;
547 continue;
550 /* partially covered */
551 final_start = max(start, ei->addr);
552 final_end = min(end, ei_end);
553 if (final_start >= final_end)
554 continue;
555 real_removed_size += final_end - final_start;
558 * left range could be head or tail, so need to update
559 * size at first.
561 ei->size -= final_end - final_start;
562 if (ei->addr < final_start)
563 continue;
564 ei->addr = final_end;
566 return real_removed_size;
569 void __init update_e820(void)
571 u32 nr_map;
573 nr_map = e820.nr_map;
574 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
575 return;
576 e820.nr_map = nr_map;
577 printk(KERN_INFO "modified physical RAM map:\n");
578 e820_print_map("modified");
580 static void __init update_e820_saved(void)
582 u32 nr_map;
584 nr_map = e820_saved.nr_map;
585 if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
586 return;
587 e820_saved.nr_map = nr_map;
589 #define MAX_GAP_END 0x100000000ull
591 * Search for a gap in the e820 memory space from start_addr to end_addr.
593 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
594 unsigned long start_addr, unsigned long long end_addr)
596 unsigned long long last;
597 int i = e820.nr_map;
598 int found = 0;
600 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
602 while (--i >= 0) {
603 unsigned long long start = e820.map[i].addr;
604 unsigned long long end = start + e820.map[i].size;
606 if (end < start_addr)
607 continue;
610 * Since "last" is at most 4GB, we know we'll
611 * fit in 32 bits if this condition is true
613 if (last > end) {
614 unsigned long gap = last - end;
616 if (gap >= *gapsize) {
617 *gapsize = gap;
618 *gapstart = end;
619 found = 1;
622 if (start < last)
623 last = start;
625 return found;
629 * Search for the biggest gap in the low 32 bits of the e820
630 * memory space. We pass this space to PCI to assign MMIO resources
631 * for hotplug or unconfigured devices in.
632 * Hopefully the BIOS let enough space left.
634 __init void e820_setup_gap(void)
636 unsigned long gapstart, gapsize;
637 int found;
639 gapstart = 0x10000000;
640 gapsize = 0x400000;
641 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
643 #ifdef CONFIG_X86_64
644 if (!found) {
645 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
646 printk(KERN_ERR
647 "PCI: Warning: Cannot find a gap in the 32bit address range\n"
648 "PCI: Unassigned devices with 32bit resource registers may break!\n");
650 #endif
653 * e820_reserve_resources_late protect stolen RAM already
655 pci_mem_start = gapstart;
657 printk(KERN_INFO
658 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
659 pci_mem_start, gapstart, gapsize);
663 * Because of the size limitation of struct boot_params, only first
664 * 128 E820 memory entries are passed to kernel via
665 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
666 * linked list of struct setup_data, which is parsed here.
668 void __init parse_e820_ext(struct setup_data *sdata, unsigned long pa_data)
670 u32 map_len;
671 int entries;
672 struct e820entry *extmap;
674 entries = sdata->len / sizeof(struct e820entry);
675 map_len = sdata->len + sizeof(struct setup_data);
676 if (map_len > PAGE_SIZE)
677 sdata = early_ioremap(pa_data, map_len);
678 extmap = (struct e820entry *)(sdata->data);
679 __append_e820_map(extmap, entries);
680 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
681 if (map_len > PAGE_SIZE)
682 early_iounmap(sdata, map_len);
683 printk(KERN_INFO "extended physical RAM map:\n");
684 e820_print_map("extended");
687 #if defined(CONFIG_X86_64) || \
688 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
690 * Find the ranges of physical addresses that do not correspond to
691 * e820 RAM areas and mark the corresponding pages as nosave for
692 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
694 * This function requires the e820 map to be sorted and without any
695 * overlapping entries and assumes the first e820 area to be RAM.
697 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
699 int i;
700 unsigned long pfn;
702 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
703 for (i = 1; i < e820.nr_map; i++) {
704 struct e820entry *ei = &e820.map[i];
706 if (pfn < PFN_UP(ei->addr))
707 register_nosave_region(pfn, PFN_UP(ei->addr));
709 pfn = PFN_DOWN(ei->addr + ei->size);
710 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
711 register_nosave_region(PFN_UP(ei->addr), pfn);
713 if (pfn >= limit_pfn)
714 break;
717 #endif
719 #ifdef CONFIG_HIBERNATION
721 * Mark ACPI NVS memory region, so that we can save/restore it during
722 * hibernation and the subsequent resume.
724 static int __init e820_mark_nvs_memory(void)
726 int i;
728 for (i = 0; i < e820.nr_map; i++) {
729 struct e820entry *ei = &e820.map[i];
731 if (ei->type == E820_NVS)
732 suspend_nvs_register(ei->addr, ei->size);
735 return 0;
737 core_initcall(e820_mark_nvs_memory);
738 #endif
741 * Find a free area with specified alignment in a specific range.
743 u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
745 int i;
747 for (i = 0; i < e820.nr_map; i++) {
748 struct e820entry *ei = &e820.map[i];
749 u64 addr;
750 u64 ei_start, ei_last;
752 if (ei->type != E820_RAM)
753 continue;
755 ei_last = ei->addr + ei->size;
756 ei_start = ei->addr;
757 addr = find_early_area(ei_start, ei_last, start, end,
758 size, align);
760 if (addr != -1ULL)
761 return addr;
763 return -1ULL;
766 u64 __init find_fw_memmap_area(u64 start, u64 end, u64 size, u64 align)
768 return find_e820_area(start, end, size, align);
771 u64 __init get_max_mapped(void)
773 u64 end = max_pfn_mapped;
775 end <<= PAGE_SHIFT;
777 return end;
780 * Find next free range after *start
782 u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
784 int i;
786 for (i = 0; i < e820.nr_map; i++) {
787 struct e820entry *ei = &e820.map[i];
788 u64 addr;
789 u64 ei_start, ei_last;
791 if (ei->type != E820_RAM)
792 continue;
794 ei_last = ei->addr + ei->size;
795 ei_start = ei->addr;
796 addr = find_early_area_size(ei_start, ei_last, start,
797 sizep, align);
799 if (addr != -1ULL)
800 return addr;
803 return -1ULL;
807 * pre allocated 4k and reserved it in e820
809 u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
811 u64 size = 0;
812 u64 addr;
813 u64 start;
815 for (start = startt; ; start += size) {
816 start = find_e820_area_size(start, &size, align);
817 if (!(start + 1))
818 return 0;
819 if (size >= sizet)
820 break;
823 #ifdef CONFIG_X86_32
824 if (start >= MAXMEM)
825 return 0;
826 if (start + size > MAXMEM)
827 size = MAXMEM - start;
828 #endif
830 addr = round_down(start + size - sizet, align);
831 if (addr < start)
832 return 0;
833 e820_update_range(addr, sizet, E820_RAM, E820_RESERVED);
834 e820_update_range_saved(addr, sizet, E820_RAM, E820_RESERVED);
835 printk(KERN_INFO "update e820 for early_reserve_e820\n");
836 update_e820();
837 update_e820_saved();
839 return addr;
842 #ifdef CONFIG_X86_32
843 # ifdef CONFIG_X86_PAE
844 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
845 # else
846 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
847 # endif
848 #else /* CONFIG_X86_32 */
849 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
850 #endif
853 * Find the highest page frame number we have available
855 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
857 int i;
858 unsigned long last_pfn = 0;
859 unsigned long max_arch_pfn = MAX_ARCH_PFN;
861 for (i = 0; i < e820.nr_map; i++) {
862 struct e820entry *ei = &e820.map[i];
863 unsigned long start_pfn;
864 unsigned long end_pfn;
866 if (ei->type != type)
867 continue;
869 start_pfn = ei->addr >> PAGE_SHIFT;
870 end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
872 if (start_pfn >= limit_pfn)
873 continue;
874 if (end_pfn > limit_pfn) {
875 last_pfn = limit_pfn;
876 break;
878 if (end_pfn > last_pfn)
879 last_pfn = end_pfn;
882 if (last_pfn > max_arch_pfn)
883 last_pfn = max_arch_pfn;
885 printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
886 last_pfn, max_arch_pfn);
887 return last_pfn;
889 unsigned long __init e820_end_of_ram_pfn(void)
891 return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
894 unsigned long __init e820_end_of_low_ram_pfn(void)
896 return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
899 * Finds an active region in the address range from start_pfn to last_pfn and
900 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
902 int __init e820_find_active_region(const struct e820entry *ei,
903 unsigned long start_pfn,
904 unsigned long last_pfn,
905 unsigned long *ei_startpfn,
906 unsigned long *ei_endpfn)
908 u64 align = PAGE_SIZE;
910 *ei_startpfn = round_up(ei->addr, align) >> PAGE_SHIFT;
911 *ei_endpfn = round_down(ei->addr + ei->size, align) >> PAGE_SHIFT;
913 /* Skip map entries smaller than a page */
914 if (*ei_startpfn >= *ei_endpfn)
915 return 0;
917 /* Skip if map is outside the node */
918 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
919 *ei_startpfn >= last_pfn)
920 return 0;
922 /* Check for overlaps */
923 if (*ei_startpfn < start_pfn)
924 *ei_startpfn = start_pfn;
925 if (*ei_endpfn > last_pfn)
926 *ei_endpfn = last_pfn;
928 return 1;
931 /* Walk the e820 map and register active regions within a node */
932 void __init e820_register_active_regions(int nid, unsigned long start_pfn,
933 unsigned long last_pfn)
935 unsigned long ei_startpfn;
936 unsigned long ei_endpfn;
937 int i;
939 for (i = 0; i < e820.nr_map; i++)
940 if (e820_find_active_region(&e820.map[i],
941 start_pfn, last_pfn,
942 &ei_startpfn, &ei_endpfn))
943 add_active_range(nid, ei_startpfn, ei_endpfn);
947 * Find the hole size (in bytes) in the memory range.
948 * @start: starting address of the memory range to scan
949 * @end: ending address of the memory range to scan
951 u64 __init e820_hole_size(u64 start, u64 end)
953 unsigned long start_pfn = start >> PAGE_SHIFT;
954 unsigned long last_pfn = end >> PAGE_SHIFT;
955 unsigned long ei_startpfn, ei_endpfn, ram = 0;
956 int i;
958 for (i = 0; i < e820.nr_map; i++) {
959 if (e820_find_active_region(&e820.map[i],
960 start_pfn, last_pfn,
961 &ei_startpfn, &ei_endpfn))
962 ram += ei_endpfn - ei_startpfn;
964 return end - start - ((u64)ram << PAGE_SHIFT);
967 static void early_panic(char *msg)
969 early_printk(msg);
970 panic(msg);
973 static int userdef __initdata;
975 /* "mem=nopentium" disables the 4MB page tables. */
976 static int __init parse_memopt(char *p)
978 u64 mem_size;
980 if (!p)
981 return -EINVAL;
983 if (!strcmp(p, "nopentium")) {
984 #ifdef CONFIG_X86_32
985 setup_clear_cpu_cap(X86_FEATURE_PSE);
986 return 0;
987 #else
988 printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
989 return -EINVAL;
990 #endif
993 userdef = 1;
994 mem_size = memparse(p, &p);
995 /* don't remove all of memory when handling "mem={invalid}" param */
996 if (mem_size == 0)
997 return -EINVAL;
998 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1000 return 0;
1002 early_param("mem", parse_memopt);
1004 static int __init parse_memmap_opt(char *p)
1006 char *oldp;
1007 u64 start_at, mem_size;
1009 if (!p)
1010 return -EINVAL;
1012 if (!strncmp(p, "exactmap", 8)) {
1013 #ifdef CONFIG_CRASH_DUMP
1015 * If we are doing a crash dump, we still need to know
1016 * the real mem size before original memory map is
1017 * reset.
1019 saved_max_pfn = e820_end_of_ram_pfn();
1020 #endif
1021 e820.nr_map = 0;
1022 userdef = 1;
1023 return 0;
1026 oldp = p;
1027 mem_size = memparse(p, &p);
1028 if (p == oldp)
1029 return -EINVAL;
1031 userdef = 1;
1032 if (*p == '@') {
1033 start_at = memparse(p+1, &p);
1034 e820_add_region(start_at, mem_size, E820_RAM);
1035 } else if (*p == '#') {
1036 start_at = memparse(p+1, &p);
1037 e820_add_region(start_at, mem_size, E820_ACPI);
1038 } else if (*p == '$') {
1039 start_at = memparse(p+1, &p);
1040 e820_add_region(start_at, mem_size, E820_RESERVED);
1041 } else
1042 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1044 return *p == '\0' ? 0 : -EINVAL;
1046 early_param("memmap", parse_memmap_opt);
1048 void __init finish_e820_parsing(void)
1050 if (userdef) {
1051 u32 nr = e820.nr_map;
1053 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
1054 early_panic("Invalid user supplied memory map");
1055 e820.nr_map = nr;
1057 printk(KERN_INFO "user-defined physical RAM map:\n");
1058 e820_print_map("user");
1062 static inline const char *e820_type_to_string(int e820_type)
1064 switch (e820_type) {
1065 case E820_RESERVED_KERN:
1066 case E820_RAM: return "System RAM";
1067 case E820_ACPI: return "ACPI Tables";
1068 case E820_NVS: return "ACPI Non-volatile Storage";
1069 case E820_UNUSABLE: return "Unusable memory";
1070 default: return "reserved";
1075 * Mark e820 reserved areas as busy for the resource manager.
1077 static struct resource __initdata *e820_res;
1078 void __init e820_reserve_resources(void)
1080 int i;
1081 struct resource *res;
1082 u64 end;
1084 res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
1085 e820_res = res;
1086 for (i = 0; i < e820.nr_map; i++) {
1087 end = e820.map[i].addr + e820.map[i].size - 1;
1088 if (end != (resource_size_t)end) {
1089 res++;
1090 continue;
1092 res->name = e820_type_to_string(e820.map[i].type);
1093 res->start = e820.map[i].addr;
1094 res->end = end;
1096 res->flags = IORESOURCE_MEM;
1099 * don't register the region that could be conflicted with
1100 * pci device BAR resource and insert them later in
1101 * pcibios_resource_survey()
1103 if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20)) {
1104 res->flags |= IORESOURCE_BUSY;
1105 insert_resource(&iomem_resource, res);
1107 res++;
1110 for (i = 0; i < e820_saved.nr_map; i++) {
1111 struct e820entry *entry = &e820_saved.map[i];
1112 firmware_map_add_early(entry->addr,
1113 entry->addr + entry->size - 1,
1114 e820_type_to_string(entry->type));
1118 /* How much should we pad RAM ending depending on where it is? */
1119 static unsigned long ram_alignment(resource_size_t pos)
1121 unsigned long mb = pos >> 20;
1123 /* To 64kB in the first megabyte */
1124 if (!mb)
1125 return 64*1024;
1127 /* To 1MB in the first 16MB */
1128 if (mb < 16)
1129 return 1024*1024;
1131 /* To 64MB for anything above that */
1132 return 64*1024*1024;
1135 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1137 void __init e820_reserve_resources_late(void)
1139 int i;
1140 struct resource *res;
1142 res = e820_res;
1143 for (i = 0; i < e820.nr_map; i++) {
1144 if (!res->parent && res->end)
1145 insert_resource_expand_to_fit(&iomem_resource, res);
1146 res++;
1150 * Try to bump up RAM regions to reasonable boundaries to
1151 * avoid stolen RAM:
1153 for (i = 0; i < e820.nr_map; i++) {
1154 struct e820entry *entry = &e820.map[i];
1155 u64 start, end;
1157 if (entry->type != E820_RAM)
1158 continue;
1159 start = entry->addr + entry->size;
1160 end = round_up(start, ram_alignment(start)) - 1;
1161 if (end > MAX_RESOURCE_SIZE)
1162 end = MAX_RESOURCE_SIZE;
1163 if (start >= end)
1164 continue;
1165 printk(KERN_DEBUG "reserve RAM buffer: %016llx - %016llx ",
1166 start, end);
1167 reserve_region_with_split(&iomem_resource, start, end,
1168 "RAM buffer");
1172 char *__init default_machine_specific_memory_setup(void)
1174 char *who = "BIOS-e820";
1175 u32 new_nr;
1177 * Try to copy the BIOS-supplied E820-map.
1179 * Otherwise fake a memory map; one section from 0k->640k,
1180 * the next section from 1mb->appropriate_mem_k
1182 new_nr = boot_params.e820_entries;
1183 sanitize_e820_map(boot_params.e820_map,
1184 ARRAY_SIZE(boot_params.e820_map),
1185 &new_nr);
1186 boot_params.e820_entries = new_nr;
1187 if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1188 < 0) {
1189 u64 mem_size;
1191 /* compare results from other methods and take the greater */
1192 if (boot_params.alt_mem_k
1193 < boot_params.screen_info.ext_mem_k) {
1194 mem_size = boot_params.screen_info.ext_mem_k;
1195 who = "BIOS-88";
1196 } else {
1197 mem_size = boot_params.alt_mem_k;
1198 who = "BIOS-e801";
1201 e820.nr_map = 0;
1202 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1203 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1206 /* In case someone cares... */
1207 return who;
1210 void __init setup_memory_map(void)
1212 char *who;
1214 who = x86_init.resources.memory_setup();
1215 memcpy(&e820_saved, &e820, sizeof(struct e820map));
1216 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1217 e820_print_map(who);