]>
Commit | Line | Data |
---|---|---|
2f36fa13 | 1 | /* |
1da177e4 LT |
2 | * Handle the memory map. |
3 | * The functions here do the job until bootmem takes over. | |
8059b2a2 VP |
4 | * |
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 | * | |
1da177e4 | 10 | */ |
1da177e4 LT |
11 | #include <linux/kernel.h> |
12 | #include <linux/types.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/bootmem.h> | |
15 | #include <linux/ioport.h> | |
16 | #include <linux/string.h> | |
5f5609df | 17 | #include <linux/kexec.h> |
b9491ac8 | 18 | #include <linux/module.h> |
e8eff5ac | 19 | #include <linux/mm.h> |
74dfd666 RW |
20 | #include <linux/suspend.h> |
21 | #include <linux/pfn.h> | |
b9491ac8 | 22 | |
1a91023a | 23 | #include <asm/pgtable.h> |
1da177e4 LT |
24 | #include <asm/page.h> |
25 | #include <asm/e820.h> | |
26 | #include <asm/proto.h> | |
30c82645 | 27 | #include <asm/setup.h> |
2bc0414e | 28 | #include <asm/sections.h> |
718fc13b | 29 | #include <asm/kdebug.h> |
1da177e4 | 30 | |
b92e9fac | 31 | struct e820map e820; |
3bd4d18c | 32 | |
2f36fa13 | 33 | /* |
1da177e4 LT |
34 | * PFN of last memory page. |
35 | */ | |
2f36fa13 | 36 | unsigned long end_pfn; |
1da177e4 | 37 | |
2f36fa13 | 38 | /* |
1da177e4 LT |
39 | * 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. | |
2f36fa13 TG |
42 | */ |
43 | unsigned long end_pfn_map; | |
1da177e4 | 44 | |
2f36fa13 | 45 | /* |
1da177e4 LT |
46 | * Last pfn which the user wants to use. |
47 | */ | |
caff0710 | 48 | static unsigned long __initdata end_user_pfn = MAXMEM>>PAGE_SHIFT; |
1da177e4 | 49 | |
75175278 AK |
50 | /* |
51 | * Early reserved memory areas. | |
52 | */ | |
53 | #define MAX_EARLY_RES 20 | |
30c82645 | 54 | |
75175278 AK |
55 | struct early_res { |
56 | unsigned long start, end; | |
25eff8d4 | 57 | char name[16]; |
75175278 AK |
58 | }; |
59 | static struct early_res early_res[MAX_EARLY_RES] __initdata = { | |
25eff8d4 | 60 | { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */ |
75175278 | 61 | #ifdef CONFIG_SMP |
25eff8d4 | 62 | { SMP_TRAMPOLINE_BASE, SMP_TRAMPOLINE_BASE + 2*PAGE_SIZE, "SMP_TRAMPOLINE" }, |
1da177e4 | 63 | #endif |
75175278 AK |
64 | {} |
65 | }; | |
66 | ||
25eff8d4 | 67 | void __init reserve_early(unsigned long start, unsigned long end, char *name) |
75175278 AK |
68 | { |
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) | |
25eff8d4 YL |
74 | 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); | |
1da177e4 | 76 | } |
75175278 AK |
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; | |
25eff8d4 YL |
82 | if (name) |
83 | strncpy(r->name, name, sizeof(r->name) - 1); | |
75175278 | 84 | } |
ac71d12c | 85 | |
75175278 AK |
86 | void __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]; | |
25eff8d4 YL |
91 | printk(KERN_INFO "early res: %d [%lx-%lx] %s\n", i, |
92 | r->start, r->end - 1, r->name); | |
75175278 | 93 | reserve_bootmem_generic(r->start, r->end - r->start); |
ac71d12c | 94 | } |
75175278 | 95 | } |
ac71d12c | 96 | |
75175278 | 97 | /* Check for already reserved areas */ |
48c508b3 YL |
98 | static inline int |
99 | bad_addr(unsigned long *addrp, unsigned long size, unsigned long align) | |
75175278 AK |
100 | { |
101 | int i; | |
102 | unsigned long addr = *addrp, last; | |
103 | int changed = 0; | |
104 | again: | |
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) { | |
48c508b3 | 109 | *addrp = addr = round_up(r->end, align); |
75175278 AK |
110 | changed = 1; |
111 | goto again; | |
112 | } | |
076422d2 | 113 | } |
75175278 | 114 | return changed; |
2f36fa13 | 115 | } |
1da177e4 | 116 | |
272b9cad YL |
117 | /* Check for already reserved areas */ |
118 | static inline int | |
119 | bad_addr_size(unsigned long *addrp, unsigned long *sizep, unsigned long align) | |
120 | { | |
121 | int i; | |
122 | unsigned long addr = *addrp, last; | |
123 | unsigned long size = *sizep; | |
124 | int changed = 0; | |
125 | again: | |
126 | last = addr + size; | |
127 | for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) { | |
128 | struct early_res *r = &early_res[i]; | |
129 | if (last > r->start && addr < r->start) { | |
130 | size = r->start - addr; | |
131 | changed = 1; | |
132 | goto again; | |
133 | } | |
134 | if (last > r->end && addr < r->end) { | |
135 | addr = round_up(r->end, align); | |
136 | size = last - addr; | |
137 | changed = 1; | |
138 | goto again; | |
139 | } | |
140 | if (last <= r->end && addr >= r->start) { | |
141 | (*sizep)++; | |
142 | return 0; | |
143 | } | |
144 | } | |
145 | if (changed) { | |
146 | *addrp = addr; | |
147 | *sizep = size; | |
148 | } | |
149 | return changed; | |
150 | } | |
95222368 AV |
151 | /* |
152 | * This function checks if any part of the range <start,end> is mapped | |
153 | * with type. | |
154 | */ | |
b92e9fac | 155 | int |
eee5a9fa | 156 | e820_any_mapped(unsigned long start, unsigned long end, unsigned type) |
2f36fa13 | 157 | { |
1da177e4 | 158 | int i; |
2f36fa13 TG |
159 | |
160 | for (i = 0; i < e820.nr_map; i++) { | |
161 | struct e820entry *ei = &e820.map[i]; | |
162 | ||
163 | if (type && ei->type != type) | |
1da177e4 | 164 | continue; |
48c8b113 | 165 | if (ei->addr >= end || ei->addr + ei->size <= start) |
2f36fa13 TG |
166 | continue; |
167 | return 1; | |
168 | } | |
1da177e4 LT |
169 | return 0; |
170 | } | |
b92e9fac | 171 | EXPORT_SYMBOL_GPL(e820_any_mapped); |
1da177e4 | 172 | |
79e453d4 LT |
173 | /* |
174 | * This function checks if the entire range <start,end> is mapped with type. | |
175 | * | |
176 | * Note: this function only works correct if the e820 table is sorted and | |
177 | * not-overlapping, which is the case | |
178 | */ | |
2f36fa13 TG |
179 | int __init e820_all_mapped(unsigned long start, unsigned long end, |
180 | unsigned type) | |
79e453d4 LT |
181 | { |
182 | int i; | |
2f36fa13 | 183 | |
79e453d4 LT |
184 | for (i = 0; i < e820.nr_map; i++) { |
185 | struct e820entry *ei = &e820.map[i]; | |
2f36fa13 | 186 | |
79e453d4 LT |
187 | if (type && ei->type != type) |
188 | continue; | |
189 | /* is the region (part) in overlap with the current region ?*/ | |
190 | if (ei->addr >= end || ei->addr + ei->size <= start) | |
191 | continue; | |
192 | ||
193 | /* if the region is at the beginning of <start,end> we move | |
194 | * start to the end of the region since it's ok until there | |
195 | */ | |
196 | if (ei->addr <= start) | |
197 | start = ei->addr + ei->size; | |
2f36fa13 TG |
198 | /* |
199 | * if start is now at or beyond end, we're done, full | |
200 | * coverage | |
201 | */ | |
79e453d4 | 202 | if (start >= end) |
2f36fa13 | 203 | return 1; |
79e453d4 LT |
204 | } |
205 | return 0; | |
206 | } | |
207 | ||
2f36fa13 | 208 | /* |
24a5da73 | 209 | * Find a free area with specified alignment in a specific range. |
2f36fa13 TG |
210 | */ |
211 | unsigned long __init find_e820_area(unsigned long start, unsigned long end, | |
48c508b3 | 212 | unsigned long size, unsigned long align) |
2f36fa13 TG |
213 | { |
214 | int i; | |
215 | ||
216 | for (i = 0; i < e820.nr_map; i++) { | |
217 | struct e820entry *ei = &e820.map[i]; | |
48c508b3 YL |
218 | unsigned long addr, last; |
219 | unsigned long ei_last; | |
2f36fa13 TG |
220 | |
221 | if (ei->type != E820_RAM) | |
222 | continue; | |
48c508b3 YL |
223 | addr = round_up(ei->addr, align); |
224 | ei_last = ei->addr + ei->size; | |
2f36fa13 | 225 | if (addr < start) |
48c508b3 | 226 | addr = round_up(start, align); |
272b9cad | 227 | if (addr >= ei_last) |
2f36fa13 | 228 | continue; |
48c508b3 | 229 | while (bad_addr(&addr, size, align) && addr+size <= ei_last) |
1da177e4 | 230 | ; |
24a5da73 | 231 | last = addr + size; |
48c508b3 | 232 | if (last > ei_last) |
1da177e4 | 233 | continue; |
2f36fa13 | 234 | if (last > end) |
1da177e4 | 235 | continue; |
2f36fa13 TG |
236 | return addr; |
237 | } | |
238 | return -1UL; | |
239 | } | |
1da177e4 | 240 | |
272b9cad YL |
241 | /* |
242 | * Find next free range after *start | |
243 | */ | |
c64df707 YL |
244 | unsigned long __init find_e820_area_size(unsigned long start, |
245 | unsigned long *sizep, | |
246 | unsigned long align) | |
272b9cad YL |
247 | { |
248 | int i; | |
249 | ||
250 | for (i = 0; i < e820.nr_map; i++) { | |
251 | struct e820entry *ei = &e820.map[i]; | |
252 | unsigned long addr, last; | |
253 | unsigned long ei_last; | |
254 | ||
255 | if (ei->type != E820_RAM) | |
256 | continue; | |
257 | addr = round_up(ei->addr, align); | |
258 | ei_last = ei->addr + ei->size; | |
272b9cad YL |
259 | if (addr < start) |
260 | addr = round_up(start, align); | |
272b9cad YL |
261 | if (addr >= ei_last) |
262 | continue; | |
263 | *sizep = ei_last - addr; | |
c64df707 YL |
264 | while (bad_addr_size(&addr, sizep, align) && |
265 | addr + *sizep <= ei_last) | |
272b9cad YL |
266 | ; |
267 | last = addr + *sizep; | |
272b9cad YL |
268 | if (last > ei_last) |
269 | continue; | |
270 | return addr; | |
271 | } | |
272 | return -1UL; | |
273 | ||
274 | } | |
1da177e4 LT |
275 | /* |
276 | * Find the highest page frame number we have available | |
277 | */ | |
278 | unsigned long __init e820_end_of_ram(void) | |
279 | { | |
2f36fa13 TG |
280 | unsigned long end_pfn; |
281 | ||
5cb248ab | 282 | end_pfn = find_max_pfn_with_active_regions(); |
2f36fa13 TG |
283 | |
284 | if (end_pfn > end_pfn_map) | |
1da177e4 LT |
285 | end_pfn_map = end_pfn; |
286 | if (end_pfn_map > MAXMEM>>PAGE_SHIFT) | |
287 | end_pfn_map = MAXMEM>>PAGE_SHIFT; | |
288 | if (end_pfn > end_user_pfn) | |
289 | end_pfn = end_user_pfn; | |
2f36fa13 TG |
290 | if (end_pfn > end_pfn_map) |
291 | end_pfn = end_pfn_map; | |
1da177e4 | 292 | |
2f36fa13 TG |
293 | printk(KERN_INFO "end_pfn_map = %lu\n", end_pfn_map); |
294 | return end_pfn; | |
1da177e4 LT |
295 | } |
296 | ||
485761bd | 297 | /* |
1da177e4 LT |
298 | * Mark e820 reserved areas as busy for the resource manager. |
299 | */ | |
3def3d6d | 300 | void __init e820_reserve_resources(void) |
1da177e4 LT |
301 | { |
302 | int i; | |
01561264 YL |
303 | struct resource *res; |
304 | ||
305 | res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map); | |
1da177e4 | 306 | for (i = 0; i < e820.nr_map; i++) { |
1da177e4 LT |
307 | switch (e820.map[i].type) { |
308 | case E820_RAM: res->name = "System RAM"; break; | |
309 | case E820_ACPI: res->name = "ACPI Tables"; break; | |
310 | case E820_NVS: res->name = "ACPI Non-volatile Storage"; break; | |
311 | default: res->name = "reserved"; | |
312 | } | |
313 | res->start = e820.map[i].addr; | |
314 | res->end = res->start + e820.map[i].size - 1; | |
315 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; | |
3def3d6d | 316 | insert_resource(&iomem_resource, res); |
01561264 | 317 | res++; |
1da177e4 LT |
318 | } |
319 | } | |
320 | ||
e8eff5ac RW |
321 | /* |
322 | * Find the ranges of physical addresses that do not correspond to | |
323 | * e820 RAM areas and mark the corresponding pages as nosave for software | |
324 | * suspend and suspend to RAM. | |
325 | * | |
326 | * This function requires the e820 map to be sorted and without any | |
327 | * overlapping entries and assumes the first e820 area to be RAM. | |
328 | */ | |
329 | void __init e820_mark_nosave_regions(void) | |
330 | { | |
331 | int i; | |
332 | unsigned long paddr; | |
333 | ||
334 | paddr = round_down(e820.map[0].addr + e820.map[0].size, PAGE_SIZE); | |
335 | for (i = 1; i < e820.nr_map; i++) { | |
336 | struct e820entry *ei = &e820.map[i]; | |
337 | ||
338 | if (paddr < ei->addr) | |
74dfd666 RW |
339 | register_nosave_region(PFN_DOWN(paddr), |
340 | PFN_UP(ei->addr)); | |
e8eff5ac RW |
341 | |
342 | paddr = round_down(ei->addr + ei->size, PAGE_SIZE); | |
343 | if (ei->type != E820_RAM) | |
74dfd666 RW |
344 | register_nosave_region(PFN_UP(ei->addr), |
345 | PFN_DOWN(paddr)); | |
e8eff5ac RW |
346 | |
347 | if (paddr >= (end_pfn << PAGE_SHIFT)) | |
348 | break; | |
349 | } | |
350 | } | |
351 | ||
3af044e0 DR |
352 | /* |
353 | * Finds an active region in the address range from start_pfn to end_pfn and | |
354 | * returns its range in ei_startpfn and ei_endpfn for the e820 entry. | |
355 | */ | |
356 | static int __init e820_find_active_region(const struct e820entry *ei, | |
357 | unsigned long start_pfn, | |
358 | unsigned long end_pfn, | |
359 | unsigned long *ei_startpfn, | |
360 | unsigned long *ei_endpfn) | |
361 | { | |
362 | *ei_startpfn = round_up(ei->addr, PAGE_SIZE) >> PAGE_SHIFT; | |
363 | *ei_endpfn = round_down(ei->addr + ei->size, PAGE_SIZE) >> PAGE_SHIFT; | |
364 | ||
365 | /* Skip map entries smaller than a page */ | |
366 | if (*ei_startpfn >= *ei_endpfn) | |
367 | return 0; | |
368 | ||
369 | /* Check if end_pfn_map should be updated */ | |
370 | if (ei->type != E820_RAM && *ei_endpfn > end_pfn_map) | |
371 | end_pfn_map = *ei_endpfn; | |
372 | ||
373 | /* Skip if map is outside the node */ | |
374 | if (ei->type != E820_RAM || *ei_endpfn <= start_pfn || | |
375 | *ei_startpfn >= end_pfn) | |
376 | return 0; | |
377 | ||
378 | /* Check for overlaps */ | |
379 | if (*ei_startpfn < start_pfn) | |
380 | *ei_startpfn = start_pfn; | |
381 | if (*ei_endpfn > end_pfn) | |
382 | *ei_endpfn = end_pfn; | |
383 | ||
384 | /* Obey end_user_pfn to save on memmap */ | |
385 | if (*ei_startpfn >= end_user_pfn) | |
386 | return 0; | |
387 | if (*ei_endpfn > end_user_pfn) | |
388 | *ei_endpfn = end_user_pfn; | |
389 | ||
390 | return 1; | |
391 | } | |
392 | ||
5cb248ab MG |
393 | /* Walk the e820 map and register active regions within a node */ |
394 | void __init | |
395 | e820_register_active_regions(int nid, unsigned long start_pfn, | |
396 | unsigned long end_pfn) | |
397 | { | |
3af044e0 DR |
398 | unsigned long ei_startpfn; |
399 | unsigned long ei_endpfn; | |
5cb248ab | 400 | int i; |
5cb248ab | 401 | |
3af044e0 DR |
402 | for (i = 0; i < e820.nr_map; i++) |
403 | if (e820_find_active_region(&e820.map[i], | |
404 | start_pfn, end_pfn, | |
405 | &ei_startpfn, &ei_endpfn)) | |
406 | add_active_range(nid, ei_startpfn, ei_endpfn); | |
5cb248ab MG |
407 | } |
408 | ||
2f36fa13 | 409 | /* |
1da177e4 | 410 | * Add a memory region to the kernel e820 map. |
2f36fa13 | 411 | */ |
1da177e4 LT |
412 | void __init add_memory_region(unsigned long start, unsigned long size, int type) |
413 | { | |
414 | int x = e820.nr_map; | |
415 | ||
416 | if (x == E820MAX) { | |
417 | printk(KERN_ERR "Ooops! Too many entries in the memory map!\n"); | |
418 | return; | |
419 | } | |
420 | ||
421 | e820.map[x].addr = start; | |
422 | e820.map[x].size = size; | |
423 | e820.map[x].type = type; | |
424 | e820.nr_map++; | |
425 | } | |
426 | ||
a7e96629 DR |
427 | /* |
428 | * Find the hole size (in bytes) in the memory range. | |
429 | * @start: starting address of the memory range to scan | |
430 | * @end: ending address of the memory range to scan | |
431 | */ | |
432 | unsigned long __init e820_hole_size(unsigned long start, unsigned long end) | |
433 | { | |
434 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
435 | unsigned long end_pfn = end >> PAGE_SHIFT; | |
2f36fa13 | 436 | unsigned long ei_startpfn, ei_endpfn, ram = 0; |
a7e96629 DR |
437 | int i; |
438 | ||
439 | for (i = 0; i < e820.nr_map; i++) { | |
440 | if (e820_find_active_region(&e820.map[i], | |
441 | start_pfn, end_pfn, | |
442 | &ei_startpfn, &ei_endpfn)) | |
443 | ram += ei_endpfn - ei_startpfn; | |
444 | } | |
445 | return end - start - (ram << PAGE_SHIFT); | |
446 | } | |
447 | ||
013d23e1 | 448 | static void __init e820_print_map(char *who) |
1da177e4 LT |
449 | { |
450 | int i; | |
451 | ||
452 | for (i = 0; i < e820.nr_map; i++) { | |
5a3ece79 | 453 | printk(KERN_INFO " %s: %016Lx - %016Lx ", who, |
2f36fa13 TG |
454 | (unsigned long long) e820.map[i].addr, |
455 | (unsigned long long) | |
456 | (e820.map[i].addr + e820.map[i].size)); | |
1da177e4 | 457 | switch (e820.map[i].type) { |
2f36fa13 TG |
458 | case E820_RAM: |
459 | printk(KERN_CONT "(usable)\n"); | |
460 | break; | |
1da177e4 | 461 | case E820_RESERVED: |
2f36fa13 TG |
462 | printk(KERN_CONT "(reserved)\n"); |
463 | break; | |
1da177e4 | 464 | case E820_ACPI: |
2f36fa13 TG |
465 | printk(KERN_CONT "(ACPI data)\n"); |
466 | break; | |
1da177e4 | 467 | case E820_NVS: |
2f36fa13 TG |
468 | printk(KERN_CONT "(ACPI NVS)\n"); |
469 | break; | |
470 | default: | |
471 | printk(KERN_CONT "type %u\n", e820.map[i].type); | |
472 | break; | |
1da177e4 LT |
473 | } |
474 | } | |
475 | } | |
476 | ||
477 | /* | |
478 | * Sanitize the BIOS e820 map. | |
479 | * | |
2f36fa13 | 480 | * Some e820 responses include overlapping entries. The following |
1da177e4 LT |
481 | * replaces the original e820 map with a new one, removing overlaps. |
482 | * | |
483 | */ | |
2f36fa13 | 484 | static int __init sanitize_e820_map(struct e820entry *biosmap, char *pnr_map) |
1da177e4 LT |
485 | { |
486 | struct change_member { | |
487 | struct e820entry *pbios; /* pointer to original bios entry */ | |
488 | unsigned long long addr; /* address for this change point */ | |
489 | }; | |
490 | static struct change_member change_point_list[2*E820MAX] __initdata; | |
491 | static struct change_member *change_point[2*E820MAX] __initdata; | |
492 | static struct e820entry *overlap_list[E820MAX] __initdata; | |
493 | static struct e820entry new_bios[E820MAX] __initdata; | |
494 | struct change_member *change_tmp; | |
495 | unsigned long current_type, last_type; | |
496 | unsigned long long last_addr; | |
497 | int chgidx, still_changing; | |
498 | int overlap_entries; | |
499 | int new_bios_entry; | |
8059b2a2 | 500 | int old_nr, new_nr, chg_nr; |
1da177e4 LT |
501 | int i; |
502 | ||
503 | /* | |
2f36fa13 TG |
504 | Visually we're performing the following |
505 | (1,2,3,4 = memory types)... | |
1da177e4 LT |
506 | |
507 | Sample memory map (w/overlaps): | |
508 | ____22__________________ | |
509 | ______________________4_ | |
510 | ____1111________________ | |
511 | _44_____________________ | |
512 | 11111111________________ | |
513 | ____________________33__ | |
514 | ___________44___________ | |
515 | __________33333_________ | |
516 | ______________22________ | |
517 | ___________________2222_ | |
518 | _________111111111______ | |
519 | _____________________11_ | |
520 | _________________4______ | |
521 | ||
522 | Sanitized equivalent (no overlap): | |
523 | 1_______________________ | |
524 | _44_____________________ | |
525 | ___1____________________ | |
526 | ____22__________________ | |
527 | ______11________________ | |
528 | _________1______________ | |
529 | __________3_____________ | |
530 | ___________44___________ | |
531 | _____________33_________ | |
532 | _______________2________ | |
533 | ________________1_______ | |
534 | _________________4______ | |
535 | ___________________2____ | |
536 | ____________________33__ | |
537 | ______________________4_ | |
538 | */ | |
539 | ||
540 | /* if there's only one memory region, don't bother */ | |
541 | if (*pnr_map < 2) | |
542 | return -1; | |
543 | ||
544 | old_nr = *pnr_map; | |
545 | ||
546 | /* bail out if we find any unreasonable addresses in bios map */ | |
2f36fa13 | 547 | for (i = 0; i < old_nr; i++) |
1da177e4 LT |
548 | if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) |
549 | return -1; | |
550 | ||
551 | /* create pointers for initial change-point information (for sorting) */ | |
2f36fa13 | 552 | for (i = 0; i < 2 * old_nr; i++) |
1da177e4 LT |
553 | change_point[i] = &change_point_list[i]; |
554 | ||
8059b2a2 VP |
555 | /* record all known change-points (starting and ending addresses), |
556 | omitting those that are for empty memory regions */ | |
1da177e4 | 557 | chgidx = 0; |
2f36fa13 | 558 | for (i = 0; i < old_nr; i++) { |
8059b2a2 VP |
559 | if (biosmap[i].size != 0) { |
560 | change_point[chgidx]->addr = biosmap[i].addr; | |
561 | change_point[chgidx++]->pbios = &biosmap[i]; | |
2f36fa13 TG |
562 | change_point[chgidx]->addr = biosmap[i].addr + |
563 | biosmap[i].size; | |
8059b2a2 VP |
564 | change_point[chgidx++]->pbios = &biosmap[i]; |
565 | } | |
1da177e4 | 566 | } |
8059b2a2 | 567 | chg_nr = chgidx; |
1da177e4 LT |
568 | |
569 | /* sort change-point list by memory addresses (low -> high) */ | |
570 | still_changing = 1; | |
571 | while (still_changing) { | |
572 | still_changing = 0; | |
2f36fa13 TG |
573 | for (i = 1; i < chg_nr; i++) { |
574 | unsigned long long curaddr, lastaddr; | |
575 | unsigned long long curpbaddr, lastpbaddr; | |
576 | ||
577 | curaddr = change_point[i]->addr; | |
578 | lastaddr = change_point[i - 1]->addr; | |
579 | curpbaddr = change_point[i]->pbios->addr; | |
580 | lastpbaddr = change_point[i - 1]->pbios->addr; | |
581 | ||
582 | /* | |
583 | * swap entries, when: | |
584 | * | |
585 | * curaddr > lastaddr or | |
586 | * curaddr == lastaddr and curaddr == curpbaddr and | |
587 | * lastaddr != lastpbaddr | |
588 | */ | |
589 | if (curaddr < lastaddr || | |
590 | (curaddr == lastaddr && curaddr == curpbaddr && | |
591 | lastaddr != lastpbaddr)) { | |
1da177e4 LT |
592 | change_tmp = change_point[i]; |
593 | change_point[i] = change_point[i-1]; | |
594 | change_point[i-1] = change_tmp; | |
2f36fa13 | 595 | still_changing = 1; |
1da177e4 LT |
596 | } |
597 | } | |
598 | } | |
599 | ||
600 | /* create a new bios memory map, removing overlaps */ | |
2f36fa13 TG |
601 | overlap_entries = 0; /* number of entries in the overlap table */ |
602 | new_bios_entry = 0; /* index for creating new bios map entries */ | |
1da177e4 LT |
603 | last_type = 0; /* start with undefined memory type */ |
604 | last_addr = 0; /* start with 0 as last starting address */ | |
2f36fa13 | 605 | |
1da177e4 | 606 | /* loop through change-points, determining affect on the new bios map */ |
2f36fa13 | 607 | for (chgidx = 0; chgidx < chg_nr; chgidx++) { |
1da177e4 | 608 | /* keep track of all overlapping bios entries */ |
2f36fa13 TG |
609 | if (change_point[chgidx]->addr == |
610 | change_point[chgidx]->pbios->addr) { | |
611 | /* | |
612 | * add map entry to overlap list (> 1 entry | |
613 | * implies an overlap) | |
614 | */ | |
615 | overlap_list[overlap_entries++] = | |
616 | change_point[chgidx]->pbios; | |
617 | } else { | |
618 | /* | |
619 | * remove entry from list (order independent, | |
620 | * so swap with last) | |
621 | */ | |
622 | for (i = 0; i < overlap_entries; i++) { | |
623 | if (overlap_list[i] == | |
624 | change_point[chgidx]->pbios) | |
625 | overlap_list[i] = | |
626 | overlap_list[overlap_entries-1]; | |
1da177e4 LT |
627 | } |
628 | overlap_entries--; | |
629 | } | |
2f36fa13 TG |
630 | /* |
631 | * if there are overlapping entries, decide which | |
632 | * "type" to use (larger value takes precedence -- | |
633 | * 1=usable, 2,3,4,4+=unusable) | |
634 | */ | |
1da177e4 | 635 | current_type = 0; |
2f36fa13 | 636 | for (i = 0; i < overlap_entries; i++) |
1da177e4 LT |
637 | if (overlap_list[i]->type > current_type) |
638 | current_type = overlap_list[i]->type; | |
2f36fa13 TG |
639 | /* |
640 | * continue building up new bios map based on this | |
641 | * information | |
642 | */ | |
1da177e4 LT |
643 | if (current_type != last_type) { |
644 | if (last_type != 0) { | |
645 | new_bios[new_bios_entry].size = | |
646 | change_point[chgidx]->addr - last_addr; | |
2f36fa13 TG |
647 | /* |
648 | * move forward only if the new size | |
649 | * was non-zero | |
650 | */ | |
1da177e4 | 651 | if (new_bios[new_bios_entry].size != 0) |
2f36fa13 TG |
652 | /* |
653 | * no more space left for new | |
654 | * bios entries ? | |
655 | */ | |
1da177e4 | 656 | if (++new_bios_entry >= E820MAX) |
2f36fa13 | 657 | break; |
1da177e4 LT |
658 | } |
659 | if (current_type != 0) { | |
2f36fa13 TG |
660 | new_bios[new_bios_entry].addr = |
661 | change_point[chgidx]->addr; | |
1da177e4 | 662 | new_bios[new_bios_entry].type = current_type; |
2f36fa13 | 663 | last_addr = change_point[chgidx]->addr; |
1da177e4 LT |
664 | } |
665 | last_type = current_type; | |
666 | } | |
667 | } | |
2f36fa13 TG |
668 | /* retain count for new bios entries */ |
669 | new_nr = new_bios_entry; | |
1da177e4 LT |
670 | |
671 | /* copy new bios mapping into original location */ | |
2f36fa13 | 672 | memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry)); |
1da177e4 LT |
673 | *pnr_map = new_nr; |
674 | ||
675 | return 0; | |
676 | } | |
677 | ||
678 | /* | |
679 | * Copy the BIOS e820 map into a safe place. | |
680 | * | |
681 | * Sanity-check it while we're at it.. | |
682 | * | |
683 | * If we're lucky and live on a modern system, the setup code | |
684 | * will have given us a memory map that we can use to properly | |
685 | * set up memory. If we aren't, we'll fake a memory map. | |
1da177e4 | 686 | */ |
2f36fa13 | 687 | static int __init copy_e820_map(struct e820entry *biosmap, int nr_map) |
1da177e4 LT |
688 | { |
689 | /* Only one memory region (or negative)? Ignore it */ | |
690 | if (nr_map < 2) | |
691 | return -1; | |
692 | ||
693 | do { | |
320a6b2e AH |
694 | u64 start = biosmap->addr; |
695 | u64 size = biosmap->size; | |
696 | u64 end = start + size; | |
697 | u32 type = biosmap->type; | |
1da177e4 LT |
698 | |
699 | /* Overflow in 64 bits? Ignore the memory map. */ | |
700 | if (start > end) | |
701 | return -1; | |
702 | ||
1da177e4 | 703 | add_memory_region(start, size, type); |
2f36fa13 | 704 | } while (biosmap++, --nr_map); |
1da177e4 LT |
705 | return 0; |
706 | } | |
707 | ||
013d23e1 | 708 | static void early_panic(char *msg) |
1da177e4 | 709 | { |
8380aabb AK |
710 | early_printk(msg); |
711 | panic(msg); | |
712 | } | |
1da177e4 | 713 | |
746ef0cd GOC |
714 | /* We're not void only for x86 32-bit compat */ |
715 | char * __init machine_specific_memory_setup(void) | |
8380aabb | 716 | { |
746ef0cd | 717 | char *who = "BIOS-e820"; |
1da177e4 LT |
718 | /* |
719 | * Try to copy the BIOS-supplied E820-map. | |
720 | * | |
721 | * Otherwise fake a memory map; one section from 0k->640k, | |
722 | * the next section from 1mb->appropriate_mem_k | |
723 | */ | |
30c82645 PA |
724 | sanitize_e820_map(boot_params.e820_map, &boot_params.e820_entries); |
725 | if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0) | |
8380aabb | 726 | early_panic("Cannot find a valid memory map"); |
1da177e4 | 727 | printk(KERN_INFO "BIOS-provided physical RAM map:\n"); |
746ef0cd GOC |
728 | e820_print_map(who); |
729 | ||
730 | /* In case someone cares... */ | |
731 | return who; | |
1da177e4 LT |
732 | } |
733 | ||
2c8c0e6b AK |
734 | static int __init parse_memopt(char *p) |
735 | { | |
736 | if (!p) | |
737 | return -EINVAL; | |
738 | end_user_pfn = memparse(p, &p); | |
2f36fa13 | 739 | end_user_pfn >>= PAGE_SHIFT; |
2c8c0e6b | 740 | return 0; |
2f36fa13 | 741 | } |
2c8c0e6b | 742 | early_param("mem", parse_memopt); |
1da177e4 | 743 | |
2c8c0e6b | 744 | static int userdef __initdata; |
1da177e4 | 745 | |
2c8c0e6b | 746 | static int __init parse_memmap_opt(char *p) |
69cda7b1 | 747 | { |
2c8c0e6b | 748 | char *oldp; |
69cda7b1 AM |
749 | unsigned long long start_at, mem_size; |
750 | ||
2c8c0e6b AK |
751 | if (!strcmp(p, "exactmap")) { |
752 | #ifdef CONFIG_CRASH_DUMP | |
2f36fa13 TG |
753 | /* |
754 | * If we are doing a crash dump, we still need to know | |
755 | * the real mem size before original memory map is | |
2c8c0e6b AK |
756 | * reset. |
757 | */ | |
15803a43 | 758 | e820_register_active_regions(0, 0, -1UL); |
2c8c0e6b | 759 | saved_max_pfn = e820_end_of_ram(); |
15803a43 | 760 | remove_all_active_ranges(); |
2c8c0e6b AK |
761 | #endif |
762 | end_pfn_map = 0; | |
763 | e820.nr_map = 0; | |
764 | userdef = 1; | |
765 | return 0; | |
766 | } | |
767 | ||
768 | oldp = p; | |
769 | mem_size = memparse(p, &p); | |
770 | if (p == oldp) | |
771 | return -EINVAL; | |
b3ca74a2 VB |
772 | |
773 | userdef = 1; | |
69cda7b1 | 774 | if (*p == '@') { |
2c8c0e6b | 775 | start_at = memparse(p+1, &p); |
69cda7b1 AM |
776 | add_memory_region(start_at, mem_size, E820_RAM); |
777 | } else if (*p == '#') { | |
2c8c0e6b | 778 | start_at = memparse(p+1, &p); |
69cda7b1 AM |
779 | add_memory_region(start_at, mem_size, E820_ACPI); |
780 | } else if (*p == '$') { | |
2c8c0e6b | 781 | start_at = memparse(p+1, &p); |
69cda7b1 AM |
782 | add_memory_region(start_at, mem_size, E820_RESERVED); |
783 | } else { | |
784 | end_user_pfn = (mem_size >> PAGE_SHIFT); | |
785 | } | |
2c8c0e6b AK |
786 | return *p == '\0' ? 0 : -EINVAL; |
787 | } | |
788 | early_param("memmap", parse_memmap_opt); | |
789 | ||
43999d9e | 790 | void __init finish_e820_parsing(void) |
2c8c0e6b AK |
791 | { |
792 | if (userdef) { | |
b3ca74a2 VB |
793 | char nr = e820.nr_map; |
794 | ||
795 | if (sanitize_e820_map(e820.map, &nr) < 0) | |
796 | early_panic("Invalid user supplied memory map"); | |
797 | e820.nr_map = nr; | |
798 | ||
2c8c0e6b AK |
799 | printk(KERN_INFO "user-defined physical RAM map:\n"); |
800 | e820_print_map("user"); | |
801 | } | |
69cda7b1 AM |
802 | } |
803 | ||
5dca6a1b YL |
804 | void __init update_memory_range(u64 start, u64 size, unsigned old_type, |
805 | unsigned new_type) | |
806 | { | |
807 | int i; | |
808 | ||
809 | BUG_ON(old_type == new_type); | |
810 | ||
811 | for (i = 0; i < e820.nr_map; i++) { | |
812 | struct e820entry *ei = &e820.map[i]; | |
813 | u64 final_start, final_end; | |
814 | if (ei->type != old_type) | |
815 | continue; | |
816 | /* totally covered? */ | |
817 | if (ei->addr >= start && ei->size <= size) { | |
818 | ei->type = new_type; | |
819 | continue; | |
820 | } | |
821 | /* partially covered */ | |
822 | final_start = max(start, ei->addr); | |
823 | final_end = min(start + size, ei->addr + ei->size); | |
824 | if (final_start >= final_end) | |
825 | continue; | |
826 | add_memory_region(final_start, final_end - final_start, | |
827 | new_type); | |
828 | } | |
829 | } | |
830 | ||
aaf23042 YL |
831 | void __init update_e820(void) |
832 | { | |
833 | u8 nr_map; | |
834 | ||
835 | nr_map = e820.nr_map; | |
836 | if (sanitize_e820_map(e820.map, &nr_map)) | |
837 | return; | |
838 | e820.nr_map = nr_map; | |
839 | printk(KERN_INFO "modified physical RAM map:\n"); | |
840 | e820_print_map("modified"); | |
841 | } | |
842 | ||
a1e97782 | 843 | unsigned long pci_mem_start = 0xaeedbabe; |
2ee60e17 | 844 | EXPORT_SYMBOL(pci_mem_start); |
a1e97782 AK |
845 | |
846 | /* | |
847 | * Search for the biggest gap in the low 32 bits of the e820 | |
848 | * memory space. We pass this space to PCI to assign MMIO resources | |
849 | * for hotplug or unconfigured devices in. | |
850 | * Hopefully the BIOS let enough space left. | |
851 | */ | |
852 | __init void e820_setup_gap(void) | |
853 | { | |
f0eca962 | 854 | unsigned long gapstart, gapsize, round; |
a1e97782 AK |
855 | unsigned long last; |
856 | int i; | |
857 | int found = 0; | |
858 | ||
859 | last = 0x100000000ull; | |
860 | gapstart = 0x10000000; | |
861 | gapsize = 0x400000; | |
862 | i = e820.nr_map; | |
863 | while (--i >= 0) { | |
864 | unsigned long long start = e820.map[i].addr; | |
865 | unsigned long long end = start + e820.map[i].size; | |
866 | ||
867 | /* | |
868 | * Since "last" is at most 4GB, we know we'll | |
869 | * fit in 32 bits if this condition is true | |
870 | */ | |
871 | if (last > end) { | |
872 | unsigned long gap = last - end; | |
873 | ||
874 | if (gap > gapsize) { | |
875 | gapsize = gap; | |
876 | gapstart = end; | |
877 | found = 1; | |
878 | } | |
879 | } | |
880 | if (start < last) | |
881 | last = start; | |
882 | } | |
883 | ||
884 | if (!found) { | |
885 | gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024; | |
2f36fa13 TG |
886 | printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit " |
887 | "address range\n" | |
888 | KERN_ERR "PCI: Unassigned devices with 32bit resource " | |
889 | "registers may break!\n"); | |
a1e97782 AK |
890 | } |
891 | ||
892 | /* | |
f0eca962 DR |
893 | * See how much we want to round up: start off with |
894 | * rounding to the next 1MB area. | |
a1e97782 | 895 | */ |
f0eca962 DR |
896 | round = 0x100000; |
897 | while ((gapsize >> 4) > round) | |
898 | round += round; | |
899 | /* Fun with two's complement */ | |
900 | pci_mem_start = (gapstart + round) & -round; | |
a1e97782 | 901 | |
2f36fa13 TG |
902 | printk(KERN_INFO |
903 | "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n", | |
904 | pci_mem_start, gapstart, gapsize); | |
a1e97782 | 905 | } |
e820482c KA |
906 | |
907 | int __init arch_get_ram_range(int slot, u64 *addr, u64 *size) | |
908 | { | |
909 | int i; | |
910 | ||
911 | if (slot < 0 || slot >= e820.nr_map) | |
912 | return -1; | |
913 | for (i = slot; i < e820.nr_map; i++) { | |
914 | if (e820.map[i].type != E820_RAM) | |
915 | continue; | |
916 | break; | |
917 | } | |
918 | if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT)) | |
919 | return -1; | |
920 | *addr = e820.map[i].addr; | |
921 | *size = min_t(u64, e820.map[i].size + e820.map[i].addr, | |
922 | max_pfn << PAGE_SHIFT) - *addr; | |
923 | return i + 1; | |
924 | } |