blob: 8b914a833ac62546e7608d44c4bd1b4c1603cad0 [file] [log] [blame]
Thomas Gleixner2f36fa12008-01-30 13:30:12 +01001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Handle the memory map.
3 * The functions here do the job until bootmem takes over.
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -07004 *
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>
9 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#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>
Eric W. Biederman5f5609d2005-06-25 14:58:04 -070017#include <linux/kexec.h>
Andrew Mortonb9491ac2005-09-16 19:27:54 -070018#include <linux/module.h>
Rafael J. Wysockie8eff5a2006-09-25 23:32:46 -070019#include <linux/mm.h>
Rafael J. Wysocki74dfd662007-05-06 14:50:43 -070020#include <linux/suspend.h>
21#include <linux/pfn.h>
Andrew Mortonb9491ac2005-09-16 19:27:54 -070022
Andrew Morton1a91023a2006-07-10 04:43:49 -070023#include <asm/pgtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <asm/page.h>
25#include <asm/e820.h>
26#include <asm/proto.h>
H. Peter Anvin30c82642007-10-15 17:13:22 -070027#include <asm/setup.h>
Andi Kleen2bc04142005-11-05 17:25:53 +010028#include <asm/sections.h>
Thomas Gleixner718fc132008-01-30 13:30:17 +010029#include <asm/kdebug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
Jan Beulichb92e9fa2007-05-02 19:27:11 +020031struct e820map e820;
Andi Kleen3bd4d182006-09-26 10:52:33 +020032
Thomas Gleixner2f36fa12008-01-30 13:30:12 +010033/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 * PFN of last memory page.
35 */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +010036unsigned long end_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Thomas Gleixner2f36fa12008-01-30 13:30:12 +010038/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 * 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.
Thomas Gleixner2f36fa12008-01-30 13:30:12 +010042 */
43unsigned long end_pfn_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Thomas Gleixner2f36fa12008-01-30 13:30:12 +010045/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 * Last pfn which the user wants to use.
47 */
Jan Beulichcaff0712006-09-26 10:52:31 +020048static unsigned long __initdata end_user_pfn = MAXMEM>>PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Andi Kleen75175272008-01-30 13:33:17 +010050/*
51 * Early reserved memory areas.
52 */
53#define MAX_EARLY_RES 20
54
55struct early_res {
56 unsigned long start, end;
Yinghai Lu25eff8d2008-02-01 17:49:41 +010057 char name[16];
Andi Kleen75175272008-01-30 13:33:17 +010058};
59static struct early_res early_res[MAX_EARLY_RES] __initdata = {
Yinghai Lu25eff8d2008-02-01 17:49:41 +010060 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
Andi Kleen75175272008-01-30 13:33:17 +010061#ifdef CONFIG_SMP
Yinghai Lu25eff8d2008-02-01 17:49:41 +010062 { SMP_TRAMPOLINE_BASE, SMP_TRAMPOLINE_BASE + 2*PAGE_SIZE, "SMP_TRAMPOLINE" },
Andi Kleen75175272008-01-30 13:33:17 +010063#endif
64 {}
65};
66
Yinghai Lu25eff8d2008-02-01 17:49:41 +010067void __init reserve_early(unsigned long start, unsigned long end, char *name)
Andi Kleen75175272008-01-30 13:33:17 +010068{
69 int i;
70 struct early_res *r;
71 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
72 r = &early_res[i];
73 if (end > r->start && start < r->end)
Yinghai Lu25eff8d2008-02-01 17:49:41 +010074 panic("Overlapping early reservations %lx-%lx %s to %lx-%lx %s\n",
75 start, end - 1, name?name:"", r->start, r->end - 1, r->name);
Andi Kleen75175272008-01-30 13:33:17 +010076 }
77 if (i >= MAX_EARLY_RES)
78 panic("Too many early reservations");
79 r = &early_res[i];
80 r->start = start;
81 r->end = end;
Yinghai Lu25eff8d2008-02-01 17:49:41 +010082 if (name)
83 strncpy(r->name, name, sizeof(r->name) - 1);
Andi Kleen75175272008-01-30 13:33:17 +010084}
85
86void __init early_res_to_bootmem(void)
87{
88 int i;
89 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
90 struct early_res *r = &early_res[i];
Yinghai Lu25eff8d2008-02-01 17:49:41 +010091 printk(KERN_INFO "early res: %d [%lx-%lx] %s\n", i,
92 r->start, r->end - 1, r->name);
Andi Kleen75175272008-01-30 13:33:17 +010093 reserve_bootmem_generic(r->start, r->end - r->start);
94 }
95}
96
97/* Check for already reserved areas */
Yinghai Lu48c508b2008-04-17 17:40:45 +020098static inline int
99bad_addr(unsigned long *addrp, unsigned long size, unsigned long align)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100100{
Andi Kleen75175272008-01-30 13:33:17 +0100101 int i;
102 unsigned long addr = *addrp, last;
103 int changed = 0;
104again:
105 last = addr + size;
106 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
107 struct early_res *r = &early_res[i];
108 if (last >= r->start && addr < r->end) {
Yinghai Lu48c508b2008-04-17 17:40:45 +0200109 *addrp = addr = round_up(r->end, align);
Andi Kleen75175272008-01-30 13:33:17 +0100110 changed = 1;
111 goto again;
H. Peter Anvin30c82642007-10-15 17:13:22 -0700112 }
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100113 }
Andi Kleen75175272008-01-30 13:33:17 +0100114 return changed;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100115}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Arjan van de Ven95222362006-04-07 19:49:27 +0200117/*
118 * This function checks if any part of the range <start,end> is mapped
119 * with type.
120 */
Jan Beulichb92e9fa2007-05-02 19:27:11 +0200121int
Arjan van de Veneee5a9f2006-04-07 19:49:24 +0200122e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100123{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 int i;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100125
126 for (i = 0; i < e820.nr_map; i++) {
127 struct e820entry *ei = &e820.map[i];
128
129 if (type && ei->type != type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 continue;
Eric W. Biederman48c8b112005-09-06 15:16:20 -0700131 if (ei->addr >= end || ei->addr + ei->size <= start)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100132 continue;
133 return 1;
134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 return 0;
136}
Jan Beulichb92e9fa2007-05-02 19:27:11 +0200137EXPORT_SYMBOL_GPL(e820_any_mapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Linus Torvalds79e453d2006-09-19 08:15:22 -0700139/*
140 * This function checks if the entire range <start,end> is mapped with type.
141 *
142 * Note: this function only works correct if the e820 table is sorted and
143 * not-overlapping, which is the case
144 */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100145int __init e820_all_mapped(unsigned long start, unsigned long end,
146 unsigned type)
Linus Torvalds79e453d2006-09-19 08:15:22 -0700147{
148 int i;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100149
Linus Torvalds79e453d2006-09-19 08:15:22 -0700150 for (i = 0; i < e820.nr_map; i++) {
151 struct e820entry *ei = &e820.map[i];
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100152
Linus Torvalds79e453d2006-09-19 08:15:22 -0700153 if (type && ei->type != type)
154 continue;
155 /* is the region (part) in overlap with the current region ?*/
156 if (ei->addr >= end || ei->addr + ei->size <= start)
157 continue;
158
159 /* if the region is at the beginning of <start,end> we move
160 * start to the end of the region since it's ok until there
161 */
162 if (ei->addr <= start)
163 start = ei->addr + ei->size;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100164 /*
165 * if start is now at or beyond end, we're done, full
166 * coverage
167 */
Linus Torvalds79e453d2006-09-19 08:15:22 -0700168 if (start >= end)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100169 return 1;
Linus Torvalds79e453d2006-09-19 08:15:22 -0700170 }
171 return 0;
172}
173
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100174/*
Yinghai Lu24a5da72008-02-01 17:49:41 +0100175 * Find a free area with specified alignment in a specific range.
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100176 */
177unsigned long __init find_e820_area(unsigned long start, unsigned long end,
Yinghai Lu48c508b2008-04-17 17:40:45 +0200178 unsigned long size, unsigned long align)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100179{
180 int i;
181
182 for (i = 0; i < e820.nr_map; i++) {
183 struct e820entry *ei = &e820.map[i];
Yinghai Lu48c508b2008-04-17 17:40:45 +0200184 unsigned long addr, last;
185 unsigned long ei_last;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100186
187 if (ei->type != E820_RAM)
188 continue;
Yinghai Lu48c508b2008-04-17 17:40:45 +0200189 addr = round_up(ei->addr, align);
190 ei_last = ei->addr + ei->size;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100191 if (addr < start)
Yinghai Lu48c508b2008-04-17 17:40:45 +0200192 addr = round_up(start, align);
193 if (addr > ei_last)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100194 continue;
Yinghai Lu48c508b2008-04-17 17:40:45 +0200195 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 ;
Yinghai Lu24a5da72008-02-01 17:49:41 +0100197 last = addr + size;
Yinghai Lu48c508b2008-04-17 17:40:45 +0200198 if (last > ei_last)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 continue;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100200 if (last > end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 continue;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100202 return addr;
203 }
204 return -1UL;
205}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207/*
208 * Find the highest page frame number we have available
209 */
210unsigned long __init e820_end_of_ram(void)
211{
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100212 unsigned long end_pfn;
213
Mel Gorman5cb248a2006-09-27 01:49:52 -0700214 end_pfn = find_max_pfn_with_active_regions();
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100215
216 if (end_pfn > end_pfn_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 end_pfn_map = end_pfn;
218 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
219 end_pfn_map = MAXMEM>>PAGE_SHIFT;
220 if (end_pfn > end_user_pfn)
221 end_pfn = end_user_pfn;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100222 if (end_pfn > end_pfn_map)
223 end_pfn = end_pfn_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100225 printk(KERN_INFO "end_pfn_map = %lu\n", end_pfn_map);
226 return end_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227}
228
Andi Kleen485761b2005-08-26 18:34:10 -0700229/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 * Mark e820 reserved areas as busy for the resource manager.
231 */
Yinghai Lu3def3d62008-02-22 17:07:16 -0800232void __init e820_reserve_resources(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
234 int i;
235 for (i = 0; i < e820.nr_map; i++) {
236 struct resource *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 res = alloc_bootmem_low(sizeof(struct resource));
238 switch (e820.map[i].type) {
239 case E820_RAM: res->name = "System RAM"; break;
240 case E820_ACPI: res->name = "ACPI Tables"; break;
241 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
242 default: res->name = "reserved";
243 }
244 res->start = e820.map[i].addr;
245 res->end = res->start + e820.map[i].size - 1;
246 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Yinghai Lu3def3d62008-02-22 17:07:16 -0800247 insert_resource(&iomem_resource, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 }
249}
250
Rafael J. Wysockie8eff5a2006-09-25 23:32:46 -0700251/*
252 * Find the ranges of physical addresses that do not correspond to
253 * e820 RAM areas and mark the corresponding pages as nosave for software
254 * suspend and suspend to RAM.
255 *
256 * This function requires the e820 map to be sorted and without any
257 * overlapping entries and assumes the first e820 area to be RAM.
258 */
259void __init e820_mark_nosave_regions(void)
260{
261 int i;
262 unsigned long paddr;
263
264 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
265 for (i = 1; i < e820.nr_map; i++) {
266 struct e820entry *ei = &e820.map[i];
267
268 if (paddr < ei->addr)
Rafael J. Wysocki74dfd662007-05-06 14:50:43 -0700269 register_nosave_region(PFN_DOWN(paddr),
270 PFN_UP(ei->addr));
Rafael J. Wysockie8eff5a2006-09-25 23:32:46 -0700271
272 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
273 if (ei->type != E820_RAM)
Rafael J. Wysocki74dfd662007-05-06 14:50:43 -0700274 register_nosave_region(PFN_UP(ei->addr),
275 PFN_DOWN(paddr));
Rafael J. Wysockie8eff5a2006-09-25 23:32:46 -0700276
277 if (paddr >= (end_pfn << PAGE_SHIFT))
278 break;
279 }
280}
281
David Rientjes3af044e2007-07-21 17:10:31 +0200282/*
283 * Finds an active region in the address range from start_pfn to end_pfn and
284 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
285 */
286static int __init e820_find_active_region(const struct e820entry *ei,
287 unsigned long start_pfn,
288 unsigned long end_pfn,
289 unsigned long *ei_startpfn,
290 unsigned long *ei_endpfn)
291{
292 *ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
293 *ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE) >> PAGE_SHIFT;
294
295 /* Skip map entries smaller than a page */
296 if (*ei_startpfn >= *ei_endpfn)
297 return 0;
298
299 /* Check if end_pfn_map should be updated */
300 if (ei->type != E820_RAM && *ei_endpfn > end_pfn_map)
301 end_pfn_map = *ei_endpfn;
302
303 /* Skip if map is outside the node */
304 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
305 *ei_startpfn >= end_pfn)
306 return 0;
307
308 /* Check for overlaps */
309 if (*ei_startpfn < start_pfn)
310 *ei_startpfn = start_pfn;
311 if (*ei_endpfn > end_pfn)
312 *ei_endpfn = end_pfn;
313
314 /* Obey end_user_pfn to save on memmap */
315 if (*ei_startpfn >= end_user_pfn)
316 return 0;
317 if (*ei_endpfn > end_user_pfn)
318 *ei_endpfn = end_user_pfn;
319
320 return 1;
321}
322
Mel Gorman5cb248a2006-09-27 01:49:52 -0700323/* Walk the e820 map and register active regions within a node */
324void __init
325e820_register_active_regions(int nid, unsigned long start_pfn,
326 unsigned long end_pfn)
327{
David Rientjes3af044e2007-07-21 17:10:31 +0200328 unsigned long ei_startpfn;
329 unsigned long ei_endpfn;
Mel Gorman5cb248a2006-09-27 01:49:52 -0700330 int i;
Mel Gorman5cb248a2006-09-27 01:49:52 -0700331
David Rientjes3af044e2007-07-21 17:10:31 +0200332 for (i = 0; i < e820.nr_map; i++)
333 if (e820_find_active_region(&e820.map[i],
334 start_pfn, end_pfn,
335 &ei_startpfn, &ei_endpfn))
336 add_active_range(nid, ei_startpfn, ei_endpfn);
Mel Gorman5cb248a2006-09-27 01:49:52 -0700337}
338
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100339/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 * Add a memory region to the kernel e820 map.
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100341 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342void __init add_memory_region(unsigned long start, unsigned long size, int type)
343{
344 int x = e820.nr_map;
345
346 if (x == E820MAX) {
347 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
348 return;
349 }
350
351 e820.map[x].addr = start;
352 e820.map[x].size = size;
353 e820.map[x].type = type;
354 e820.nr_map++;
355}
356
David Rientjesa7e96622007-07-21 17:11:29 +0200357/*
358 * Find the hole size (in bytes) in the memory range.
359 * @start: starting address of the memory range to scan
360 * @end: ending address of the memory range to scan
361 */
362unsigned long __init e820_hole_size(unsigned long start, unsigned long end)
363{
364 unsigned long start_pfn = start >> PAGE_SHIFT;
365 unsigned long end_pfn = end >> PAGE_SHIFT;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100366 unsigned long ei_startpfn, ei_endpfn, ram = 0;
David Rientjesa7e96622007-07-21 17:11:29 +0200367 int i;
368
369 for (i = 0; i < e820.nr_map; i++) {
370 if (e820_find_active_region(&e820.map[i],
371 start_pfn, end_pfn,
372 &ei_startpfn, &ei_endpfn))
373 ram += ei_endpfn - ei_startpfn;
374 }
375 return end - start - (ram << PAGE_SHIFT);
376}
377
Adrian Bunk013d23e2008-01-30 13:30:30 +0100378static void __init e820_print_map(char *who)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
380 int i;
381
382 for (i = 0; i < e820.nr_map; i++) {
Dan Aloni5a3ece72007-07-21 17:11:37 +0200383 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100384 (unsigned long long) e820.map[i].addr,
385 (unsigned long long)
386 (e820.map[i].addr + e820.map[i].size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 switch (e820.map[i].type) {
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100388 case E820_RAM:
389 printk(KERN_CONT "(usable)\n");
390 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 case E820_RESERVED:
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100392 printk(KERN_CONT "(reserved)\n");
393 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 case E820_ACPI:
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100395 printk(KERN_CONT "(ACPI data)\n");
396 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 case E820_NVS:
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100398 printk(KERN_CONT "(ACPI NVS)\n");
399 break;
400 default:
401 printk(KERN_CONT "type %u\n", e820.map[i].type);
402 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 }
404 }
405}
406
407/*
408 * Sanitize the BIOS e820 map.
409 *
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100410 * Some e820 responses include overlapping entries. The following
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 * replaces the original e820 map with a new one, removing overlaps.
412 *
413 */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100414static int __init sanitize_e820_map(struct e820entry *biosmap, char *pnr_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
416 struct change_member {
417 struct e820entry *pbios; /* pointer to original bios entry */
418 unsigned long long addr; /* address for this change point */
419 };
420 static struct change_member change_point_list[2*E820MAX] __initdata;
421 static struct change_member *change_point[2*E820MAX] __initdata;
422 static struct e820entry *overlap_list[E820MAX] __initdata;
423 static struct e820entry new_bios[E820MAX] __initdata;
424 struct change_member *change_tmp;
425 unsigned long current_type, last_type;
426 unsigned long long last_addr;
427 int chgidx, still_changing;
428 int overlap_entries;
429 int new_bios_entry;
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700430 int old_nr, new_nr, chg_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 int i;
432
433 /*
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100434 Visually we're performing the following
435 (1,2,3,4 = memory types)...
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
437 Sample memory map (w/overlaps):
438 ____22__________________
439 ______________________4_
440 ____1111________________
441 _44_____________________
442 11111111________________
443 ____________________33__
444 ___________44___________
445 __________33333_________
446 ______________22________
447 ___________________2222_
448 _________111111111______
449 _____________________11_
450 _________________4______
451
452 Sanitized equivalent (no overlap):
453 1_______________________
454 _44_____________________
455 ___1____________________
456 ____22__________________
457 ______11________________
458 _________1______________
459 __________3_____________
460 ___________44___________
461 _____________33_________
462 _______________2________
463 ________________1_______
464 _________________4______
465 ___________________2____
466 ____________________33__
467 ______________________4_
468 */
469
470 /* if there's only one memory region, don't bother */
471 if (*pnr_map < 2)
472 return -1;
473
474 old_nr = *pnr_map;
475
476 /* bail out if we find any unreasonable addresses in bios map */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100477 for (i = 0; i < old_nr; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
479 return -1;
480
481 /* create pointers for initial change-point information (for sorting) */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100482 for (i = 0; i < 2 * old_nr; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 change_point[i] = &change_point_list[i];
484
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700485 /* record all known change-points (starting and ending addresses),
486 omitting those that are for empty memory regions */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 chgidx = 0;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100488 for (i = 0; i < old_nr; i++) {
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700489 if (biosmap[i].size != 0) {
490 change_point[chgidx]->addr = biosmap[i].addr;
491 change_point[chgidx++]->pbios = &biosmap[i];
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100492 change_point[chgidx]->addr = biosmap[i].addr +
493 biosmap[i].size;
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700494 change_point[chgidx++]->pbios = &biosmap[i];
495 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 }
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700497 chg_nr = chgidx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
499 /* sort change-point list by memory addresses (low -> high) */
500 still_changing = 1;
501 while (still_changing) {
502 still_changing = 0;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100503 for (i = 1; i < chg_nr; i++) {
504 unsigned long long curaddr, lastaddr;
505 unsigned long long curpbaddr, lastpbaddr;
506
507 curaddr = change_point[i]->addr;
508 lastaddr = change_point[i - 1]->addr;
509 curpbaddr = change_point[i]->pbios->addr;
510 lastpbaddr = change_point[i - 1]->pbios->addr;
511
512 /*
513 * swap entries, when:
514 *
515 * curaddr > lastaddr or
516 * curaddr == lastaddr and curaddr == curpbaddr and
517 * lastaddr != lastpbaddr
518 */
519 if (curaddr < lastaddr ||
520 (curaddr == lastaddr && curaddr == curpbaddr &&
521 lastaddr != lastpbaddr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 change_tmp = change_point[i];
523 change_point[i] = change_point[i-1];
524 change_point[i-1] = change_tmp;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100525 still_changing = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 }
527 }
528 }
529
530 /* create a new bios memory map, removing overlaps */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100531 overlap_entries = 0; /* number of entries in the overlap table */
532 new_bios_entry = 0; /* index for creating new bios map entries */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 last_type = 0; /* start with undefined memory type */
534 last_addr = 0; /* start with 0 as last starting address */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 /* loop through change-points, determining affect on the new bios map */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100537 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 /* keep track of all overlapping bios entries */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100539 if (change_point[chgidx]->addr ==
540 change_point[chgidx]->pbios->addr) {
541 /*
542 * add map entry to overlap list (> 1 entry
543 * implies an overlap)
544 */
545 overlap_list[overlap_entries++] =
546 change_point[chgidx]->pbios;
547 } else {
548 /*
549 * remove entry from list (order independent,
550 * so swap with last)
551 */
552 for (i = 0; i < overlap_entries; i++) {
553 if (overlap_list[i] ==
554 change_point[chgidx]->pbios)
555 overlap_list[i] =
556 overlap_list[overlap_entries-1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 }
558 overlap_entries--;
559 }
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100560 /*
561 * if there are overlapping entries, decide which
562 * "type" to use (larger value takes precedence --
563 * 1=usable, 2,3,4,4+=unusable)
564 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 current_type = 0;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100566 for (i = 0; i < overlap_entries; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 if (overlap_list[i]->type > current_type)
568 current_type = overlap_list[i]->type;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100569 /*
570 * continue building up new bios map based on this
571 * information
572 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 if (current_type != last_type) {
574 if (last_type != 0) {
575 new_bios[new_bios_entry].size =
576 change_point[chgidx]->addr - last_addr;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100577 /*
578 * move forward only if the new size
579 * was non-zero
580 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (new_bios[new_bios_entry].size != 0)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100582 /*
583 * no more space left for new
584 * bios entries ?
585 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (++new_bios_entry >= E820MAX)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100587 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 }
589 if (current_type != 0) {
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100590 new_bios[new_bios_entry].addr =
591 change_point[chgidx]->addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 new_bios[new_bios_entry].type = current_type;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100593 last_addr = change_point[chgidx]->addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595 last_type = current_type;
596 }
597 }
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100598 /* retain count for new bios entries */
599 new_nr = new_bios_entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601 /* copy new bios mapping into original location */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100602 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 *pnr_map = new_nr;
604
605 return 0;
606}
607
608/*
609 * Copy the BIOS e820 map into a safe place.
610 *
611 * Sanity-check it while we're at it..
612 *
613 * If we're lucky and live on a modern system, the setup code
614 * will have given us a memory map that we can use to properly
615 * set up memory. If we aren't, we'll fake a memory map.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100617static int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618{
619 /* Only one memory region (or negative)? Ignore it */
620 if (nr_map < 2)
621 return -1;
622
623 do {
624 unsigned long start = biosmap->addr;
625 unsigned long size = biosmap->size;
626 unsigned long end = start + size;
627 unsigned long type = biosmap->type;
628
629 /* Overflow in 64 bits? Ignore the memory map. */
630 if (start > end)
631 return -1;
632
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 add_memory_region(start, size, type);
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100634 } while (biosmap++, --nr_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 return 0;
636}
637
Adrian Bunk013d23e2008-01-30 13:30:30 +0100638static void early_panic(char *msg)
Andi Kleen8380aab2006-09-26 10:52:37 +0200639{
640 early_printk(msg);
641 panic(msg);
642}
643
Glauber de Oliveira Costa746ef0c2008-01-30 13:31:11 +0100644/* We're not void only for x86 32-bit compat */
645char * __init machine_specific_memory_setup(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
Glauber de Oliveira Costa746ef0c2008-01-30 13:31:11 +0100647 char *who = "BIOS-e820";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 /*
649 * Try to copy the BIOS-supplied E820-map.
650 *
651 * Otherwise fake a memory map; one section from 0k->640k,
652 * the next section from 1mb->appropriate_mem_k
653 */
H. Peter Anvin30c82642007-10-15 17:13:22 -0700654 sanitize_e820_map(boot_params.e820_map, &boot_params.e820_entries);
655 if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0)
Andi Kleen8380aab2006-09-26 10:52:37 +0200656 early_panic("Cannot find a valid memory map");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
Glauber de Oliveira Costa746ef0c2008-01-30 13:31:11 +0100658 e820_print_map(who);
659
660 /* In case someone cares... */
661 return who;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662}
663
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200664static int __init parse_memopt(char *p)
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800665{
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200666 if (!p)
667 return -EINVAL;
668 end_user_pfn = memparse(p, &p);
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100669 end_user_pfn >>= PAGE_SHIFT;
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200670 return 0;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100671}
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200672early_param("mem", parse_memopt);
673
674static int userdef __initdata;
675
676static int __init parse_memmap_opt(char *p)
677{
678 char *oldp;
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800679 unsigned long long start_at, mem_size;
680
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200681 if (!strcmp(p, "exactmap")) {
682#ifdef CONFIG_CRASH_DUMP
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100683 /*
684 * If we are doing a crash dump, we still need to know
685 * the real mem size before original memory map is
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200686 * reset.
687 */
Magnus Damm15803a42006-11-14 16:57:46 +0100688 e820_register_active_regions(0, 0, -1UL);
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200689 saved_max_pfn = e820_end_of_ram();
Magnus Damm15803a42006-11-14 16:57:46 +0100690 remove_all_active_ranges();
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200691#endif
692 end_pfn_map = 0;
693 e820.nr_map = 0;
694 userdef = 1;
695 return 0;
696 }
697
698 oldp = p;
699 mem_size = memparse(p, &p);
700 if (p == oldp)
701 return -EINVAL;
Vladimir Bereznikerb3ca74a2008-01-30 13:30:46 +0100702
703 userdef = 1;
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800704 if (*p == '@') {
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200705 start_at = memparse(p+1, &p);
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800706 add_memory_region(start_at, mem_size, E820_RAM);
707 } else if (*p == '#') {
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200708 start_at = memparse(p+1, &p);
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800709 add_memory_region(start_at, mem_size, E820_ACPI);
710 } else if (*p == '$') {
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200711 start_at = memparse(p+1, &p);
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800712 add_memory_region(start_at, mem_size, E820_RESERVED);
713 } else {
714 end_user_pfn = (mem_size >> PAGE_SHIFT);
715 }
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200716 return *p == '\0' ? 0 : -EINVAL;
717}
718early_param("memmap", parse_memmap_opt);
719
Sam Ravnborg43999d92007-03-16 21:07:36 +0100720void __init finish_e820_parsing(void)
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200721{
722 if (userdef) {
Vladimir Bereznikerb3ca74a2008-01-30 13:30:46 +0100723 char nr = e820.nr_map;
724
725 if (sanitize_e820_map(e820.map, &nr) < 0)
726 early_panic("Invalid user supplied memory map");
727 e820.nr_map = nr;
728
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200729 printk(KERN_INFO "user-defined physical RAM map:\n");
730 e820_print_map("user");
731 }
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800732}
733
Yinghai Lu5dca6a12008-03-18 16:44:19 -0700734void __init update_memory_range(u64 start, u64 size, unsigned old_type,
735 unsigned new_type)
736{
737 int i;
738
739 BUG_ON(old_type == new_type);
740
741 for (i = 0; i < e820.nr_map; i++) {
742 struct e820entry *ei = &e820.map[i];
743 u64 final_start, final_end;
744 if (ei->type != old_type)
745 continue;
746 /* totally covered? */
747 if (ei->addr >= start && ei->size <= size) {
748 ei->type = new_type;
749 continue;
750 }
751 /* partially covered */
752 final_start = max(start, ei->addr);
753 final_end = min(start + size, ei->addr + ei->size);
754 if (final_start >= final_end)
755 continue;
756 add_memory_region(final_start, final_end - final_start,
757 new_type);
758 }
759}
760
Yinghai Luaaf23042008-01-30 13:33:09 +0100761void __init update_e820(void)
762{
763 u8 nr_map;
764
765 nr_map = e820.nr_map;
766 if (sanitize_e820_map(e820.map, &nr_map))
767 return;
768 e820.nr_map = nr_map;
769 printk(KERN_INFO "modified physical RAM map:\n");
770 e820_print_map("modified");
771}
772
Andi Kleena1e97782005-04-16 15:25:12 -0700773unsigned long pci_mem_start = 0xaeedbabe;
Andi Kleen2ee60e172006-06-26 13:59:44 +0200774EXPORT_SYMBOL(pci_mem_start);
Andi Kleena1e97782005-04-16 15:25:12 -0700775
776/*
777 * Search for the biggest gap in the low 32 bits of the e820
778 * memory space. We pass this space to PCI to assign MMIO resources
779 * for hotplug or unconfigured devices in.
780 * Hopefully the BIOS let enough space left.
781 */
782__init void e820_setup_gap(void)
783{
Daniel Ritzf0eca962005-09-09 00:57:14 +0200784 unsigned long gapstart, gapsize, round;
Andi Kleena1e97782005-04-16 15:25:12 -0700785 unsigned long last;
786 int i;
787 int found = 0;
788
789 last = 0x100000000ull;
790 gapstart = 0x10000000;
791 gapsize = 0x400000;
792 i = e820.nr_map;
793 while (--i >= 0) {
794 unsigned long long start = e820.map[i].addr;
795 unsigned long long end = start + e820.map[i].size;
796
797 /*
798 * Since "last" is at most 4GB, we know we'll
799 * fit in 32 bits if this condition is true
800 */
801 if (last > end) {
802 unsigned long gap = last - end;
803
804 if (gap > gapsize) {
805 gapsize = gap;
806 gapstart = end;
807 found = 1;
808 }
809 }
810 if (start < last)
811 last = start;
812 }
813
814 if (!found) {
815 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100816 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
817 "address range\n"
818 KERN_ERR "PCI: Unassigned devices with 32bit resource "
819 "registers may break!\n");
Andi Kleena1e97782005-04-16 15:25:12 -0700820 }
821
822 /*
Daniel Ritzf0eca962005-09-09 00:57:14 +0200823 * See how much we want to round up: start off with
824 * rounding to the next 1MB area.
Andi Kleena1e97782005-04-16 15:25:12 -0700825 */
Daniel Ritzf0eca962005-09-09 00:57:14 +0200826 round = 0x100000;
827 while ((gapsize >> 4) > round)
828 round += round;
829 /* Fun with two's complement */
830 pci_mem_start = (gapstart + round) & -round;
Andi Kleena1e97782005-04-16 15:25:12 -0700831
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100832 printk(KERN_INFO
833 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
834 pci_mem_start, gapstart, gapsize);
Andi Kleena1e97782005-04-16 15:25:12 -0700835}
Keshavamurthy, Anil Se8204822007-10-21 16:41:55 -0700836
837int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
838{
839 int i;
840
841 if (slot < 0 || slot >= e820.nr_map)
842 return -1;
843 for (i = slot; i < e820.nr_map; i++) {
844 if (e820.map[i].type != E820_RAM)
845 continue;
846 break;
847 }
848 if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
849 return -1;
850 *addr = e820.map[i].addr;
851 *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
852 max_pfn << PAGE_SHIFT) - *addr;
853 return i + 1;
854}