blob: b74e83b214cc4656d38cbd929ffc10a3bc39ad9f [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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070098static inline int bad_addr(unsigned long *addrp, unsigned long size)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +010099{
Andi Kleen75175272008-01-30 13:33:17 +0100100 int i;
101 unsigned long addr = *addrp, last;
102 int changed = 0;
103again:
104 last = addr + size;
105 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
106 struct early_res *r = &early_res[i];
107 if (last >= r->start && addr < r->end) {
108 *addrp = addr = r->end;
109 changed = 1;
110 goto again;
H. Peter Anvin30c82642007-10-15 17:13:22 -0700111 }
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100112 }
Andi Kleen75175272008-01-30 13:33:17 +0100113 return changed;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100114}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Arjan van de Ven95222362006-04-07 19:49:27 +0200116/*
117 * This function checks if any part of the range <start,end> is mapped
118 * with type.
119 */
Jan Beulichb92e9fa2007-05-02 19:27:11 +0200120int
Arjan van de Veneee5a9f2006-04-07 19:49:24 +0200121e820_any_mapped(unsigned long start, unsigned long end, unsigned type)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100122{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 int i;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100124
125 for (i = 0; i < e820.nr_map; i++) {
126 struct e820entry *ei = &e820.map[i];
127
128 if (type && ei->type != type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 continue;
Eric W. Biederman48c8b112005-09-06 15:16:20 -0700130 if (ei->addr >= end || ei->addr + ei->size <= start)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100131 continue;
132 return 1;
133 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 return 0;
135}
Jan Beulichb92e9fa2007-05-02 19:27:11 +0200136EXPORT_SYMBOL_GPL(e820_any_mapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
Linus Torvalds79e453d2006-09-19 08:15:22 -0700138/*
139 * This function checks if the entire range <start,end> is mapped with type.
140 *
141 * Note: this function only works correct if the e820 table is sorted and
142 * not-overlapping, which is the case
143 */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100144int __init e820_all_mapped(unsigned long start, unsigned long end,
145 unsigned type)
Linus Torvalds79e453d2006-09-19 08:15:22 -0700146{
147 int i;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100148
Linus Torvalds79e453d2006-09-19 08:15:22 -0700149 for (i = 0; i < e820.nr_map; i++) {
150 struct e820entry *ei = &e820.map[i];
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100151
Linus Torvalds79e453d2006-09-19 08:15:22 -0700152 if (type && ei->type != type)
153 continue;
154 /* is the region (part) in overlap with the current region ?*/
155 if (ei->addr >= end || ei->addr + ei->size <= start)
156 continue;
157
158 /* if the region is at the beginning of <start,end> we move
159 * start to the end of the region since it's ok until there
160 */
161 if (ei->addr <= start)
162 start = ei->addr + ei->size;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100163 /*
164 * if start is now at or beyond end, we're done, full
165 * coverage
166 */
Linus Torvalds79e453d2006-09-19 08:15:22 -0700167 if (start >= end)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100168 return 1;
Linus Torvalds79e453d2006-09-19 08:15:22 -0700169 }
170 return 0;
171}
172
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100173/*
174 * Find a free area in a specific range.
175 */
176unsigned long __init find_e820_area(unsigned long start, unsigned long end,
177 unsigned size)
178{
179 int i;
180
181 for (i = 0; i < e820.nr_map; i++) {
182 struct e820entry *ei = &e820.map[i];
183 unsigned long addr = ei->addr, last;
184
185 if (ei->type != E820_RAM)
186 continue;
187 if (addr < start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 addr = start;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100189 if (addr > ei->addr + ei->size)
190 continue;
Robert Hentosh7ca97c62006-05-30 22:48:00 +0200191 while (bad_addr(&addr, size) && addr+size <= ei->addr+ei->size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 ;
Vivek Goyal73bb8912006-10-21 18:37:01 +0200193 last = PAGE_ALIGN(addr) + size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 if (last > ei->addr + ei->size)
195 continue;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100196 if (last > end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 continue;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100198 return addr;
199 }
200 return -1UL;
201}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203/*
204 * Find the highest page frame number we have available
205 */
206unsigned long __init e820_end_of_ram(void)
207{
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100208 unsigned long end_pfn;
209
Mel Gorman5cb248a2006-09-27 01:49:52 -0700210 end_pfn = find_max_pfn_with_active_regions();
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100211
212 if (end_pfn > end_pfn_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 end_pfn_map = end_pfn;
214 if (end_pfn_map > MAXMEM>>PAGE_SHIFT)
215 end_pfn_map = MAXMEM>>PAGE_SHIFT;
216 if (end_pfn > end_user_pfn)
217 end_pfn = end_user_pfn;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100218 if (end_pfn > end_pfn_map)
219 end_pfn = end_pfn_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100221 printk(KERN_INFO "end_pfn_map = %lu\n", end_pfn_map);
222 return end_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
Andi Kleen485761b2005-08-26 18:34:10 -0700225/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 * Mark e820 reserved areas as busy for the resource manager.
227 */
Bernhard Wallec9cce832008-01-30 13:30:32 +0100228void __init e820_reserve_resources(struct resource *code_resource,
229 struct resource *data_resource, struct resource *bss_resource)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
231 int i;
232 for (i = 0; i < e820.nr_map; i++) {
233 struct resource *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 res = alloc_bootmem_low(sizeof(struct resource));
235 switch (e820.map[i].type) {
236 case E820_RAM: res->name = "System RAM"; break;
237 case E820_ACPI: res->name = "ACPI Tables"; break;
238 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
239 default: res->name = "reserved";
240 }
241 res->start = e820.map[i].addr;
242 res->end = res->start + e820.map[i].size - 1;
243 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
244 request_resource(&iomem_resource, res);
245 if (e820.map[i].type == E820_RAM) {
246 /*
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100247 * We don't know which RAM region contains kernel data,
248 * so we try it repeatedly and let the resource manager
249 * test it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 */
Bernhard Wallec9cce832008-01-30 13:30:32 +0100251 request_resource(res, code_resource);
252 request_resource(res, data_resource);
253 request_resource(res, bss_resource);
Eric W. Biederman5f5609d2005-06-25 14:58:04 -0700254#ifdef CONFIG_KEXEC
Bernhard Walle5c3391f2007-10-18 23:40:59 -0700255 if (crashk_res.start != crashk_res.end)
256 request_resource(res, &crashk_res);
Eric W. Biederman5f5609d2005-06-25 14:58:04 -0700257#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 }
259 }
260}
261
Rafael J. Wysockie8eff5a2006-09-25 23:32:46 -0700262/*
263 * Find the ranges of physical addresses that do not correspond to
264 * e820 RAM areas and mark the corresponding pages as nosave for software
265 * suspend and suspend to RAM.
266 *
267 * This function requires the e820 map to be sorted and without any
268 * overlapping entries and assumes the first e820 area to be RAM.
269 */
270void __init e820_mark_nosave_regions(void)
271{
272 int i;
273 unsigned long paddr;
274
275 paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE);
276 for (i = 1; i < e820.nr_map; i++) {
277 struct e820entry *ei = &e820.map[i];
278
279 if (paddr < ei->addr)
Rafael J. Wysocki74dfd662007-05-06 14:50:43 -0700280 register_nosave_region(PFN_DOWN(paddr),
281 PFN_UP(ei->addr));
Rafael J. Wysockie8eff5a2006-09-25 23:32:46 -0700282
283 paddr = round_down(ei->addr + ei->size, PAGE_SIZE);
284 if (ei->type != E820_RAM)
Rafael J. Wysocki74dfd662007-05-06 14:50:43 -0700285 register_nosave_region(PFN_UP(ei->addr),
286 PFN_DOWN(paddr));
Rafael J. Wysockie8eff5a2006-09-25 23:32:46 -0700287
288 if (paddr >= (end_pfn << PAGE_SHIFT))
289 break;
290 }
291}
292
David Rientjes3af044e2007-07-21 17:10:31 +0200293/*
294 * Finds an active region in the address range from start_pfn to end_pfn and
295 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
296 */
297static int __init e820_find_active_region(const struct e820entry *ei,
298 unsigned long start_pfn,
299 unsigned long end_pfn,
300 unsigned long *ei_startpfn,
301 unsigned long *ei_endpfn)
302{
303 *ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT;
304 *ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE) >> PAGE_SHIFT;
305
306 /* Skip map entries smaller than a page */
307 if (*ei_startpfn >= *ei_endpfn)
308 return 0;
309
310 /* Check if end_pfn_map should be updated */
311 if (ei->type != E820_RAM && *ei_endpfn > end_pfn_map)
312 end_pfn_map = *ei_endpfn;
313
314 /* Skip if map is outside the node */
315 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
316 *ei_startpfn >= end_pfn)
317 return 0;
318
319 /* Check for overlaps */
320 if (*ei_startpfn < start_pfn)
321 *ei_startpfn = start_pfn;
322 if (*ei_endpfn > end_pfn)
323 *ei_endpfn = end_pfn;
324
325 /* Obey end_user_pfn to save on memmap */
326 if (*ei_startpfn >= end_user_pfn)
327 return 0;
328 if (*ei_endpfn > end_user_pfn)
329 *ei_endpfn = end_user_pfn;
330
331 return 1;
332}
333
Mel Gorman5cb248a2006-09-27 01:49:52 -0700334/* Walk the e820 map and register active regions within a node */
335void __init
336e820_register_active_regions(int nid, unsigned long start_pfn,
337 unsigned long end_pfn)
338{
David Rientjes3af044e2007-07-21 17:10:31 +0200339 unsigned long ei_startpfn;
340 unsigned long ei_endpfn;
Mel Gorman5cb248a2006-09-27 01:49:52 -0700341 int i;
Mel Gorman5cb248a2006-09-27 01:49:52 -0700342
David Rientjes3af044e2007-07-21 17:10:31 +0200343 for (i = 0; i < e820.nr_map; i++)
344 if (e820_find_active_region(&e820.map[i],
345 start_pfn, end_pfn,
346 &ei_startpfn, &ei_endpfn))
347 add_active_range(nid, ei_startpfn, ei_endpfn);
Mel Gorman5cb248a2006-09-27 01:49:52 -0700348}
349
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100350/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 * Add a memory region to the kernel e820 map.
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100352 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353void __init add_memory_region(unsigned long start, unsigned long size, int type)
354{
355 int x = e820.nr_map;
356
357 if (x == E820MAX) {
358 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
359 return;
360 }
361
362 e820.map[x].addr = start;
363 e820.map[x].size = size;
364 e820.map[x].type = type;
365 e820.nr_map++;
366}
367
David Rientjesa7e96622007-07-21 17:11:29 +0200368/*
369 * Find the hole size (in bytes) in the memory range.
370 * @start: starting address of the memory range to scan
371 * @end: ending address of the memory range to scan
372 */
373unsigned long __init e820_hole_size(unsigned long start, unsigned long end)
374{
375 unsigned long start_pfn = start >> PAGE_SHIFT;
376 unsigned long end_pfn = end >> PAGE_SHIFT;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100377 unsigned long ei_startpfn, ei_endpfn, ram = 0;
David Rientjesa7e96622007-07-21 17:11:29 +0200378 int i;
379
380 for (i = 0; i < e820.nr_map; i++) {
381 if (e820_find_active_region(&e820.map[i],
382 start_pfn, end_pfn,
383 &ei_startpfn, &ei_endpfn))
384 ram += ei_endpfn - ei_startpfn;
385 }
386 return end - start - (ram << PAGE_SHIFT);
387}
388
Adrian Bunk013d23e2008-01-30 13:30:30 +0100389static void __init e820_print_map(char *who)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
391 int i;
392
393 for (i = 0; i < e820.nr_map; i++) {
Dan Aloni5a3ece72007-07-21 17:11:37 +0200394 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100395 (unsigned long long) e820.map[i].addr,
396 (unsigned long long)
397 (e820.map[i].addr + e820.map[i].size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 switch (e820.map[i].type) {
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100399 case E820_RAM:
400 printk(KERN_CONT "(usable)\n");
401 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 case E820_RESERVED:
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100403 printk(KERN_CONT "(reserved)\n");
404 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 case E820_ACPI:
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100406 printk(KERN_CONT "(ACPI data)\n");
407 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 case E820_NVS:
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100409 printk(KERN_CONT "(ACPI NVS)\n");
410 break;
411 default:
412 printk(KERN_CONT "type %u\n", e820.map[i].type);
413 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 }
415 }
416}
417
418/*
419 * Sanitize the BIOS e820 map.
420 *
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100421 * Some e820 responses include overlapping entries. The following
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 * replaces the original e820 map with a new one, removing overlaps.
423 *
424 */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100425static int __init sanitize_e820_map(struct e820entry *biosmap, char *pnr_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426{
427 struct change_member {
428 struct e820entry *pbios; /* pointer to original bios entry */
429 unsigned long long addr; /* address for this change point */
430 };
431 static struct change_member change_point_list[2*E820MAX] __initdata;
432 static struct change_member *change_point[2*E820MAX] __initdata;
433 static struct e820entry *overlap_list[E820MAX] __initdata;
434 static struct e820entry new_bios[E820MAX] __initdata;
435 struct change_member *change_tmp;
436 unsigned long current_type, last_type;
437 unsigned long long last_addr;
438 int chgidx, still_changing;
439 int overlap_entries;
440 int new_bios_entry;
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700441 int old_nr, new_nr, chg_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 int i;
443
444 /*
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100445 Visually we're performing the following
446 (1,2,3,4 = memory types)...
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
448 Sample memory map (w/overlaps):
449 ____22__________________
450 ______________________4_
451 ____1111________________
452 _44_____________________
453 11111111________________
454 ____________________33__
455 ___________44___________
456 __________33333_________
457 ______________22________
458 ___________________2222_
459 _________111111111______
460 _____________________11_
461 _________________4______
462
463 Sanitized equivalent (no overlap):
464 1_______________________
465 _44_____________________
466 ___1____________________
467 ____22__________________
468 ______11________________
469 _________1______________
470 __________3_____________
471 ___________44___________
472 _____________33_________
473 _______________2________
474 ________________1_______
475 _________________4______
476 ___________________2____
477 ____________________33__
478 ______________________4_
479 */
480
481 /* if there's only one memory region, don't bother */
482 if (*pnr_map < 2)
483 return -1;
484
485 old_nr = *pnr_map;
486
487 /* bail out if we find any unreasonable addresses in bios map */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100488 for (i = 0; i < old_nr; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
490 return -1;
491
492 /* create pointers for initial change-point information (for sorting) */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100493 for (i = 0; i < 2 * old_nr; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 change_point[i] = &change_point_list[i];
495
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700496 /* record all known change-points (starting and ending addresses),
497 omitting those that are for empty memory regions */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 chgidx = 0;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100499 for (i = 0; i < old_nr; i++) {
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700500 if (biosmap[i].size != 0) {
501 change_point[chgidx]->addr = biosmap[i].addr;
502 change_point[chgidx++]->pbios = &biosmap[i];
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100503 change_point[chgidx]->addr = biosmap[i].addr +
504 biosmap[i].size;
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700505 change_point[chgidx++]->pbios = &biosmap[i];
506 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 }
Venkatesh Pallipadi8059b2a2005-05-01 08:58:52 -0700508 chg_nr = chgidx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510 /* sort change-point list by memory addresses (low -> high) */
511 still_changing = 1;
512 while (still_changing) {
513 still_changing = 0;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100514 for (i = 1; i < chg_nr; i++) {
515 unsigned long long curaddr, lastaddr;
516 unsigned long long curpbaddr, lastpbaddr;
517
518 curaddr = change_point[i]->addr;
519 lastaddr = change_point[i - 1]->addr;
520 curpbaddr = change_point[i]->pbios->addr;
521 lastpbaddr = change_point[i - 1]->pbios->addr;
522
523 /*
524 * swap entries, when:
525 *
526 * curaddr > lastaddr or
527 * curaddr == lastaddr and curaddr == curpbaddr and
528 * lastaddr != lastpbaddr
529 */
530 if (curaddr < lastaddr ||
531 (curaddr == lastaddr && curaddr == curpbaddr &&
532 lastaddr != lastpbaddr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 change_tmp = change_point[i];
534 change_point[i] = change_point[i-1];
535 change_point[i-1] = change_tmp;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100536 still_changing = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 }
538 }
539 }
540
541 /* create a new bios memory map, removing overlaps */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100542 overlap_entries = 0; /* number of entries in the overlap table */
543 new_bios_entry = 0; /* index for creating new bios map entries */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 last_type = 0; /* start with undefined memory type */
545 last_addr = 0; /* start with 0 as last starting address */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100546
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 /* loop through change-points, determining affect on the new bios map */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100548 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 /* keep track of all overlapping bios entries */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100550 if (change_point[chgidx]->addr ==
551 change_point[chgidx]->pbios->addr) {
552 /*
553 * add map entry to overlap list (> 1 entry
554 * implies an overlap)
555 */
556 overlap_list[overlap_entries++] =
557 change_point[chgidx]->pbios;
558 } else {
559 /*
560 * remove entry from list (order independent,
561 * so swap with last)
562 */
563 for (i = 0; i < overlap_entries; i++) {
564 if (overlap_list[i] ==
565 change_point[chgidx]->pbios)
566 overlap_list[i] =
567 overlap_list[overlap_entries-1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
569 overlap_entries--;
570 }
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100571 /*
572 * if there are overlapping entries, decide which
573 * "type" to use (larger value takes precedence --
574 * 1=usable, 2,3,4,4+=unusable)
575 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 current_type = 0;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100577 for (i = 0; i < overlap_entries; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 if (overlap_list[i]->type > current_type)
579 current_type = overlap_list[i]->type;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100580 /*
581 * continue building up new bios map based on this
582 * information
583 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 if (current_type != last_type) {
585 if (last_type != 0) {
586 new_bios[new_bios_entry].size =
587 change_point[chgidx]->addr - last_addr;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100588 /*
589 * move forward only if the new size
590 * was non-zero
591 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 if (new_bios[new_bios_entry].size != 0)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100593 /*
594 * no more space left for new
595 * bios entries ?
596 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 if (++new_bios_entry >= E820MAX)
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100598 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 }
600 if (current_type != 0) {
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100601 new_bios[new_bios_entry].addr =
602 change_point[chgidx]->addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 new_bios[new_bios_entry].type = current_type;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100604 last_addr = change_point[chgidx]->addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 }
606 last_type = current_type;
607 }
608 }
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100609 /* retain count for new bios entries */
610 new_nr = new_bios_entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 /* copy new bios mapping into original location */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100613 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 *pnr_map = new_nr;
615
616 return 0;
617}
618
619/*
620 * Copy the BIOS e820 map into a safe place.
621 *
622 * Sanity-check it while we're at it..
623 *
624 * If we're lucky and live on a modern system, the setup code
625 * will have given us a memory map that we can use to properly
626 * set up memory. If we aren't, we'll fake a memory map.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 */
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100628static int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
630 /* Only one memory region (or negative)? Ignore it */
631 if (nr_map < 2)
632 return -1;
633
634 do {
635 unsigned long start = biosmap->addr;
636 unsigned long size = biosmap->size;
637 unsigned long end = start + size;
638 unsigned long type = biosmap->type;
639
640 /* Overflow in 64 bits? Ignore the memory map. */
641 if (start > end)
642 return -1;
643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 add_memory_region(start, size, type);
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100645 } while (biosmap++, --nr_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 return 0;
647}
648
Adrian Bunk013d23e2008-01-30 13:30:30 +0100649static void early_panic(char *msg)
Andi Kleen8380aab2006-09-26 10:52:37 +0200650{
651 early_printk(msg);
652 panic(msg);
653}
654
Glauber de Oliveira Costa746ef0c2008-01-30 13:31:11 +0100655/* We're not void only for x86 32-bit compat */
656char * __init machine_specific_memory_setup(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
Glauber de Oliveira Costa746ef0c2008-01-30 13:31:11 +0100658 char *who = "BIOS-e820";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 /*
660 * Try to copy the BIOS-supplied E820-map.
661 *
662 * Otherwise fake a memory map; one section from 0k->640k,
663 * the next section from 1mb->appropriate_mem_k
664 */
H. Peter Anvin30c82642007-10-15 17:13:22 -0700665 sanitize_e820_map(boot_params.e820_map, &boot_params.e820_entries);
666 if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0)
Andi Kleen8380aab2006-09-26 10:52:37 +0200667 early_panic("Cannot find a valid memory map");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
Glauber de Oliveira Costa746ef0c2008-01-30 13:31:11 +0100669 e820_print_map(who);
670
671 /* In case someone cares... */
672 return who;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673}
674
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200675static int __init parse_memopt(char *p)
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800676{
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200677 if (!p)
678 return -EINVAL;
679 end_user_pfn = memparse(p, &p);
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100680 end_user_pfn >>= PAGE_SHIFT;
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200681 return 0;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100682}
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200683early_param("mem", parse_memopt);
684
685static int userdef __initdata;
686
687static int __init parse_memmap_opt(char *p)
688{
689 char *oldp;
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800690 unsigned long long start_at, mem_size;
691
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200692 if (!strcmp(p, "exactmap")) {
693#ifdef CONFIG_CRASH_DUMP
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100694 /*
695 * If we are doing a crash dump, we still need to know
696 * the real mem size before original memory map is
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200697 * reset.
698 */
Magnus Damm15803a42006-11-14 16:57:46 +0100699 e820_register_active_regions(0, 0, -1UL);
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200700 saved_max_pfn = e820_end_of_ram();
Magnus Damm15803a42006-11-14 16:57:46 +0100701 remove_all_active_ranges();
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200702#endif
703 end_pfn_map = 0;
704 e820.nr_map = 0;
705 userdef = 1;
706 return 0;
707 }
708
709 oldp = p;
710 mem_size = memparse(p, &p);
711 if (p == oldp)
712 return -EINVAL;
Vladimir Bereznikerb3ca74a2008-01-30 13:30:46 +0100713
714 userdef = 1;
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800715 if (*p == '@') {
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200716 start_at = memparse(p+1, &p);
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800717 add_memory_region(start_at, mem_size, E820_RAM);
718 } else if (*p == '#') {
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200719 start_at = memparse(p+1, &p);
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800720 add_memory_region(start_at, mem_size, E820_ACPI);
721 } else if (*p == '$') {
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200722 start_at = memparse(p+1, &p);
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800723 add_memory_region(start_at, mem_size, E820_RESERVED);
724 } else {
725 end_user_pfn = (mem_size >> PAGE_SHIFT);
726 }
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200727 return *p == '\0' ? 0 : -EINVAL;
728}
729early_param("memmap", parse_memmap_opt);
730
Sam Ravnborg43999d92007-03-16 21:07:36 +0100731void __init finish_e820_parsing(void)
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200732{
733 if (userdef) {
Vladimir Bereznikerb3ca74a2008-01-30 13:30:46 +0100734 char nr = e820.nr_map;
735
736 if (sanitize_e820_map(e820.map, &nr) < 0)
737 early_panic("Invalid user supplied memory map");
738 e820.nr_map = nr;
739
Andi Kleen2c8c0e62006-09-26 10:52:32 +0200740 printk(KERN_INFO "user-defined physical RAM map:\n");
741 e820_print_map("user");
742 }
akpm@osdl.org69cda7b2006-01-09 20:51:46 -0800743}
744
Yinghai Luaaf23042008-01-30 13:33:09 +0100745void __init update_e820(void)
746{
747 u8 nr_map;
748
749 nr_map = e820.nr_map;
750 if (sanitize_e820_map(e820.map, &nr_map))
751 return;
752 e820.nr_map = nr_map;
753 printk(KERN_INFO "modified physical RAM map:\n");
754 e820_print_map("modified");
755}
756
Andi Kleena1e97782005-04-16 15:25:12 -0700757unsigned long pci_mem_start = 0xaeedbabe;
Andi Kleen2ee60e172006-06-26 13:59:44 +0200758EXPORT_SYMBOL(pci_mem_start);
Andi Kleena1e97782005-04-16 15:25:12 -0700759
760/*
761 * Search for the biggest gap in the low 32 bits of the e820
762 * memory space. We pass this space to PCI to assign MMIO resources
763 * for hotplug or unconfigured devices in.
764 * Hopefully the BIOS let enough space left.
765 */
766__init void e820_setup_gap(void)
767{
Daniel Ritzf0eca962005-09-09 00:57:14 +0200768 unsigned long gapstart, gapsize, round;
Andi Kleena1e97782005-04-16 15:25:12 -0700769 unsigned long last;
770 int i;
771 int found = 0;
772
773 last = 0x100000000ull;
774 gapstart = 0x10000000;
775 gapsize = 0x400000;
776 i = e820.nr_map;
777 while (--i >= 0) {
778 unsigned long long start = e820.map[i].addr;
779 unsigned long long end = start + e820.map[i].size;
780
781 /*
782 * Since "last" is at most 4GB, we know we'll
783 * fit in 32 bits if this condition is true
784 */
785 if (last > end) {
786 unsigned long gap = last - end;
787
788 if (gap > gapsize) {
789 gapsize = gap;
790 gapstart = end;
791 found = 1;
792 }
793 }
794 if (start < last)
795 last = start;
796 }
797
798 if (!found) {
799 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100800 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
801 "address range\n"
802 KERN_ERR "PCI: Unassigned devices with 32bit resource "
803 "registers may break!\n");
Andi Kleena1e97782005-04-16 15:25:12 -0700804 }
805
806 /*
Daniel Ritzf0eca962005-09-09 00:57:14 +0200807 * See how much we want to round up: start off with
808 * rounding to the next 1MB area.
Andi Kleena1e97782005-04-16 15:25:12 -0700809 */
Daniel Ritzf0eca962005-09-09 00:57:14 +0200810 round = 0x100000;
811 while ((gapsize >> 4) > round)
812 round += round;
813 /* Fun with two's complement */
814 pci_mem_start = (gapstart + round) & -round;
Andi Kleena1e97782005-04-16 15:25:12 -0700815
Thomas Gleixner2f36fa12008-01-30 13:30:12 +0100816 printk(KERN_INFO
817 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
818 pci_mem_start, gapstart, gapsize);
Andi Kleena1e97782005-04-16 15:25:12 -0700819}
Keshavamurthy, Anil Se8204822007-10-21 16:41:55 -0700820
821int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
822{
823 int i;
824
825 if (slot < 0 || slot >= e820.nr_map)
826 return -1;
827 for (i = slot; i < e820.nr_map; i++) {
828 if (e820.map[i].type != E820_RAM)
829 continue;
830 break;
831 }
832 if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
833 return -1;
834 *addr = e820.map[i].addr;
835 *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
836 max_pfn << PAGE_SHIFT) - *addr;
837 return i + 1;
838}