Commit | Line | Data |
---|---|---|
b79cd8f1 YL |
1 | /* |
2 | * Handle the memory map. | |
3 | * The functions here do the job until bootmem takes over. | |
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 | * | |
10 | */ | |
11 | #include <linux/kernel.h> | |
12 | #include <linux/types.h> | |
13 | #include <linux/init.h> | |
93a72052 | 14 | #include <linux/crash_dump.h> |
69c60c88 | 15 | #include <linux/export.h> |
b79cd8f1 | 16 | #include <linux/bootmem.h> |
b79cd8f1 | 17 | #include <linux/pfn.h> |
bf62f398 | 18 | #include <linux/suspend.h> |
976513db | 19 | #include <linux/acpi.h> |
5dfcf14d | 20 | #include <linux/firmware-map.h> |
72d7c3b3 | 21 | #include <linux/memblock.h> |
d1bbdd66 | 22 | #include <linux/sort.h> |
b79cd8f1 | 23 | |
b79cd8f1 | 24 | #include <asm/e820.h> |
a4c81cf6 | 25 | #include <asm/proto.h> |
b79cd8f1 | 26 | #include <asm/setup.h> |
cd4d09ec | 27 | #include <asm/cpufeature.h> |
b79cd8f1 | 28 | |
5dfcf14d BW |
29 | /* |
30 | * The e820 map is the map that gets modified e.g. with command line parameters | |
31 | * and that is also registered with modifications in the kernel resource tree | |
32 | * with the iomem_resource as parent. | |
33 | * | |
34 | * The e820_saved is directly saved after the BIOS-provided memory map is | |
35 | * copied. It doesn't get modified afterwards. It's registered for the | |
36 | * /sys/firmware/memmap interface. | |
37 | * | |
38 | * That memory map is not modified and is used as base for kexec. The kexec'd | |
39 | * kernel should get the same memory map as the firmware provides. Then the | |
40 | * user can e.g. boot the original kernel with mem=1G while still booting the | |
41 | * next kernel with full memory. | |
42 | */ | |
18278229 DV |
43 | static struct e820map initial_e820 __initdata; |
44 | static struct e820map initial_e820_saved __initdata; | |
45 | struct e820map *e820 __refdata = &initial_e820; | |
46 | struct e820map *e820_saved __refdata = &initial_e820_saved; | |
b79cd8f1 YL |
47 | |
48 | /* For PCI or other memory-mapped resources */ | |
49 | unsigned long pci_mem_start = 0xaeedbabe; | |
50 | #ifdef CONFIG_PCI | |
51 | EXPORT_SYMBOL(pci_mem_start); | |
52 | #endif | |
53 | ||
54 | /* | |
55 | * This function checks if any part of the range <start,end> is mapped | |
56 | * with type. | |
57 | */ | |
58 | int | |
59 | e820_any_mapped(u64 start, u64 end, unsigned type) | |
60 | { | |
61 | int i; | |
62 | ||
47533968 DV |
63 | for (i = 0; i < e820->nr_map; i++) { |
64 | struct e820entry *ei = &e820->map[i]; | |
b79cd8f1 YL |
65 | |
66 | if (type && ei->type != type) | |
67 | continue; | |
68 | if (ei->addr >= end || ei->addr + ei->size <= start) | |
69 | continue; | |
70 | return 1; | |
71 | } | |
72 | return 0; | |
73 | } | |
74 | EXPORT_SYMBOL_GPL(e820_any_mapped); | |
75 | ||
76 | /* | |
77 | * This function checks if the entire range <start,end> is mapped with type. | |
78 | * | |
79 | * Note: this function only works correct if the e820 table is sorted and | |
80 | * not-overlapping, which is the case | |
81 | */ | |
82 | int __init e820_all_mapped(u64 start, u64 end, unsigned type) | |
83 | { | |
84 | int i; | |
85 | ||
47533968 DV |
86 | for (i = 0; i < e820->nr_map; i++) { |
87 | struct e820entry *ei = &e820->map[i]; | |
b79cd8f1 YL |
88 | |
89 | if (type && ei->type != type) | |
90 | continue; | |
91 | /* is the region (part) in overlap with the current region ?*/ | |
92 | if (ei->addr >= end || ei->addr + ei->size <= start) | |
93 | continue; | |
94 | ||
95 | /* if the region is at the beginning of <start,end> we move | |
96 | * start to the end of the region since it's ok until there | |
97 | */ | |
98 | if (ei->addr <= start) | |
99 | start = ei->addr + ei->size; | |
100 | /* | |
101 | * if start is now at or beyond end, we're done, full | |
102 | * coverage | |
103 | */ | |
104 | if (start >= end) | |
105 | return 1; | |
106 | } | |
107 | return 0; | |
108 | } | |
109 | ||
110 | /* | |
111 | * Add a memory region to the kernel e820 map. | |
112 | */ | |
773e673d YL |
113 | static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size, |
114 | int type) | |
b79cd8f1 | 115 | { |
773e673d | 116 | int x = e820x->nr_map; |
b79cd8f1 | 117 | |
5051fd69 | 118 | if (x >= ARRAY_SIZE(e820x->map)) { |
91eb0f67 BH |
119 | printk(KERN_ERR "e820: too many entries; ignoring [mem %#010llx-%#010llx]\n", |
120 | (unsigned long long) start, | |
121 | (unsigned long long) (start + size - 1)); | |
b79cd8f1 YL |
122 | return; |
123 | } | |
124 | ||
773e673d YL |
125 | e820x->map[x].addr = start; |
126 | e820x->map[x].size = size; | |
127 | e820x->map[x].type = type; | |
128 | e820x->nr_map++; | |
129 | } | |
130 | ||
131 | void __init e820_add_region(u64 start, u64 size, int type) | |
132 | { | |
47533968 | 133 | __e820_add_region(e820, start, size, type); |
b79cd8f1 YL |
134 | } |
135 | ||
c61cf4cf YL |
136 | static void __init e820_print_type(u32 type) |
137 | { | |
138 | switch (type) { | |
139 | case E820_RAM: | |
140 | case E820_RESERVED_KERN: | |
91eb0f67 | 141 | printk(KERN_CONT "usable"); |
c61cf4cf YL |
142 | break; |
143 | case E820_RESERVED: | |
91eb0f67 | 144 | printk(KERN_CONT "reserved"); |
c61cf4cf YL |
145 | break; |
146 | case E820_ACPI: | |
91eb0f67 | 147 | printk(KERN_CONT "ACPI data"); |
c61cf4cf YL |
148 | break; |
149 | case E820_NVS: | |
91eb0f67 | 150 | printk(KERN_CONT "ACPI NVS"); |
c61cf4cf YL |
151 | break; |
152 | case E820_UNUSABLE: | |
91eb0f67 | 153 | printk(KERN_CONT "unusable"); |
c61cf4cf | 154 | break; |
ad5fb870 | 155 | case E820_PMEM: |
ec776ef6 CH |
156 | case E820_PRAM: |
157 | printk(KERN_CONT "persistent (type %u)", type); | |
158 | break; | |
c61cf4cf YL |
159 | default: |
160 | printk(KERN_CONT "type %u", type); | |
161 | break; | |
162 | } | |
163 | } | |
164 | ||
b79cd8f1 YL |
165 | void __init e820_print_map(char *who) |
166 | { | |
167 | int i; | |
168 | ||
47533968 | 169 | for (i = 0; i < e820->nr_map; i++) { |
91eb0f67 | 170 | printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who, |
47533968 | 171 | (unsigned long long) e820->map[i].addr, |
b79cd8f1 | 172 | (unsigned long long) |
47533968 DV |
173 | (e820->map[i].addr + e820->map[i].size - 1)); |
174 | e820_print_type(e820->map[i].type); | |
c61cf4cf | 175 | printk(KERN_CONT "\n"); |
b79cd8f1 YL |
176 | } |
177 | } | |
178 | ||
179 | /* | |
180 | * Sanitize the BIOS e820 map. | |
181 | * | |
182 | * Some e820 responses include overlapping entries. The following | |
5b7eb2e9 PJ |
183 | * replaces the original e820 map with a new one, removing overlaps, |
184 | * and resolving conflicting memory types in favor of highest | |
185 | * numbered type. | |
b79cd8f1 | 186 | * |
5b7eb2e9 PJ |
187 | * The input parameter biosmap points to an array of 'struct |
188 | * e820entry' which on entry has elements in the range [0, *pnr_map) | |
189 | * valid, and which has space for up to max_nr_map entries. | |
190 | * On return, the resulting sanitized e820 map entries will be in | |
191 | * overwritten in the same location, starting at biosmap. | |
192 | * | |
193 | * The integer pointed to by pnr_map must be valid on entry (the | |
d574ffa1 WC |
194 | * current number of valid entries located at biosmap). If the |
195 | * sanitizing succeeds the *pnr_map will be updated with the new | |
196 | * number of valid entries (something no more than max_nr_map). | |
5b7eb2e9 PJ |
197 | * |
198 | * The return value from sanitize_e820_map() is zero if it | |
199 | * successfully 'sanitized' the map entries passed in, and is -1 | |
200 | * if it did nothing, which can happen if either of (1) it was | |
201 | * only passed one map entry, or (2) any of the input map entries | |
202 | * were invalid (start + size < start, meaning that the size was | |
203 | * so big the described memory range wrapped around through zero.) | |
204 | * | |
205 | * Visually we're performing the following | |
206 | * (1,2,3,4 = memory types)... | |
207 | * | |
208 | * Sample memory map (w/overlaps): | |
209 | * ____22__________________ | |
210 | * ______________________4_ | |
211 | * ____1111________________ | |
212 | * _44_____________________ | |
213 | * 11111111________________ | |
214 | * ____________________33__ | |
215 | * ___________44___________ | |
216 | * __________33333_________ | |
217 | * ______________22________ | |
218 | * ___________________2222_ | |
219 | * _________111111111______ | |
220 | * _____________________11_ | |
221 | * _________________4______ | |
222 | * | |
223 | * Sanitized equivalent (no overlap): | |
224 | * 1_______________________ | |
225 | * _44_____________________ | |
226 | * ___1____________________ | |
227 | * ____22__________________ | |
228 | * ______11________________ | |
229 | * _________1______________ | |
230 | * __________3_____________ | |
231 | * ___________44___________ | |
232 | * _____________33_________ | |
233 | * _______________2________ | |
234 | * ________________1_______ | |
235 | * _________________4______ | |
236 | * ___________________2____ | |
237 | * ____________________33__ | |
238 | * ______________________4_ | |
b79cd8f1 | 239 | */ |
d1bbdd66 MD |
240 | struct change_member { |
241 | struct e820entry *pbios; /* pointer to original bios entry */ | |
242 | unsigned long long addr; /* address for this change point */ | |
243 | }; | |
244 | ||
245 | static int __init cpcompare(const void *a, const void *b) | |
246 | { | |
247 | struct change_member * const *app = a, * const *bpp = b; | |
248 | const struct change_member *ap = *app, *bp = *bpp; | |
249 | ||
250 | /* | |
251 | * Inputs are pointers to two elements of change_point[]. If their | |
252 | * addresses are unequal, their difference dominates. If the addresses | |
253 | * are equal, then consider one that represents the end of its region | |
254 | * to be greater than one that does not. | |
255 | */ | |
256 | if (ap->addr != bp->addr) | |
257 | return ap->addr > bp->addr ? 1 : -1; | |
258 | ||
259 | return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr); | |
260 | } | |
5b7eb2e9 | 261 | |
c3965bd1 | 262 | int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map, |
ba639039 | 263 | u32 *pnr_map) |
b79cd8f1 | 264 | { |
157fabf0 PJ |
265 | static struct change_member change_point_list[2*E820_X_MAX] __initdata; |
266 | static struct change_member *change_point[2*E820_X_MAX] __initdata; | |
267 | static struct e820entry *overlap_list[E820_X_MAX] __initdata; | |
268 | static struct e820entry new_bios[E820_X_MAX] __initdata; | |
b79cd8f1 YL |
269 | unsigned long current_type, last_type; |
270 | unsigned long long last_addr; | |
d1bbdd66 | 271 | int chgidx; |
b79cd8f1 YL |
272 | int overlap_entries; |
273 | int new_bios_entry; | |
274 | int old_nr, new_nr, chg_nr; | |
275 | int i; | |
276 | ||
b79cd8f1 YL |
277 | /* if there's only one memory region, don't bother */ |
278 | if (*pnr_map < 2) | |
279 | return -1; | |
280 | ||
281 | old_nr = *pnr_map; | |
6e9bcc79 | 282 | BUG_ON(old_nr > max_nr_map); |
b79cd8f1 YL |
283 | |
284 | /* bail out if we find any unreasonable addresses in bios map */ | |
285 | for (i = 0; i < old_nr; i++) | |
286 | if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) | |
287 | return -1; | |
288 | ||
289 | /* create pointers for initial change-point information (for sorting) */ | |
290 | for (i = 0; i < 2 * old_nr; i++) | |
291 | change_point[i] = &change_point_list[i]; | |
292 | ||
293 | /* record all known change-points (starting and ending addresses), | |
294 | omitting those that are for empty memory regions */ | |
295 | chgidx = 0; | |
296 | for (i = 0; i < old_nr; i++) { | |
297 | if (biosmap[i].size != 0) { | |
298 | change_point[chgidx]->addr = biosmap[i].addr; | |
299 | change_point[chgidx++]->pbios = &biosmap[i]; | |
300 | change_point[chgidx]->addr = biosmap[i].addr + | |
301 | biosmap[i].size; | |
302 | change_point[chgidx++]->pbios = &biosmap[i]; | |
303 | } | |
304 | } | |
305 | chg_nr = chgidx; | |
306 | ||
307 | /* sort change-point list by memory addresses (low -> high) */ | |
706d9a9c | 308 | sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL); |
b79cd8f1 YL |
309 | |
310 | /* create a new bios memory map, removing overlaps */ | |
311 | overlap_entries = 0; /* number of entries in the overlap table */ | |
312 | new_bios_entry = 0; /* index for creating new bios map entries */ | |
313 | last_type = 0; /* start with undefined memory type */ | |
314 | last_addr = 0; /* start with 0 as last starting address */ | |
315 | ||
316 | /* loop through change-points, determining affect on the new bios map */ | |
317 | for (chgidx = 0; chgidx < chg_nr; chgidx++) { | |
318 | /* keep track of all overlapping bios entries */ | |
319 | if (change_point[chgidx]->addr == | |
320 | change_point[chgidx]->pbios->addr) { | |
321 | /* | |
322 | * add map entry to overlap list (> 1 entry | |
323 | * implies an overlap) | |
324 | */ | |
325 | overlap_list[overlap_entries++] = | |
326 | change_point[chgidx]->pbios; | |
327 | } else { | |
328 | /* | |
329 | * remove entry from list (order independent, | |
330 | * so swap with last) | |
331 | */ | |
332 | for (i = 0; i < overlap_entries; i++) { | |
333 | if (overlap_list[i] == | |
334 | change_point[chgidx]->pbios) | |
335 | overlap_list[i] = | |
336 | overlap_list[overlap_entries-1]; | |
337 | } | |
338 | overlap_entries--; | |
339 | } | |
340 | /* | |
341 | * if there are overlapping entries, decide which | |
342 | * "type" to use (larger value takes precedence -- | |
343 | * 1=usable, 2,3,4,4+=unusable) | |
344 | */ | |
345 | current_type = 0; | |
346 | for (i = 0; i < overlap_entries; i++) | |
347 | if (overlap_list[i]->type > current_type) | |
348 | current_type = overlap_list[i]->type; | |
349 | /* | |
350 | * continue building up new bios map based on this | |
351 | * information | |
352 | */ | |
917db484 | 353 | if (current_type != last_type) { |
b79cd8f1 YL |
354 | if (last_type != 0) { |
355 | new_bios[new_bios_entry].size = | |
356 | change_point[chgidx]->addr - last_addr; | |
357 | /* | |
358 | * move forward only if the new size | |
359 | * was non-zero | |
360 | */ | |
361 | if (new_bios[new_bios_entry].size != 0) | |
362 | /* | |
363 | * no more space left for new | |
364 | * bios entries ? | |
365 | */ | |
c3965bd1 | 366 | if (++new_bios_entry >= max_nr_map) |
b79cd8f1 YL |
367 | break; |
368 | } | |
369 | if (current_type != 0) { | |
370 | new_bios[new_bios_entry].addr = | |
371 | change_point[chgidx]->addr; | |
372 | new_bios[new_bios_entry].type = current_type; | |
373 | last_addr = change_point[chgidx]->addr; | |
374 | } | |
375 | last_type = current_type; | |
376 | } | |
377 | } | |
378 | /* retain count for new bios entries */ | |
379 | new_nr = new_bios_entry; | |
380 | ||
381 | /* copy new bios mapping into original location */ | |
382 | memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry)); | |
383 | *pnr_map = new_nr; | |
384 | ||
385 | return 0; | |
386 | } | |
387 | ||
dc8e8120 | 388 | static int __init __append_e820_map(struct e820entry *biosmap, int nr_map) |
8c5beb50 HY |
389 | { |
390 | while (nr_map) { | |
391 | u64 start = biosmap->addr; | |
392 | u64 size = biosmap->size; | |
3ec97965 | 393 | u64 end = start + size - 1; |
8c5beb50 HY |
394 | u32 type = biosmap->type; |
395 | ||
396 | /* Overflow in 64 bits? Ignore the memory map. */ | |
3ec97965 | 397 | if (start > end && likely(size)) |
8c5beb50 HY |
398 | return -1; |
399 | ||
400 | e820_add_region(start, size, type); | |
401 | ||
402 | biosmap++; | |
403 | nr_map--; | |
404 | } | |
405 | return 0; | |
406 | } | |
407 | ||
b79cd8f1 YL |
408 | /* |
409 | * Copy the BIOS e820 map into a safe place. | |
410 | * | |
411 | * Sanity-check it while we're at it.. | |
412 | * | |
413 | * If we're lucky and live on a modern system, the setup code | |
414 | * will have given us a memory map that we can use to properly | |
415 | * set up memory. If we aren't, we'll fake a memory map. | |
416 | */ | |
dc8e8120 | 417 | static int __init append_e820_map(struct e820entry *biosmap, int nr_map) |
b79cd8f1 YL |
418 | { |
419 | /* Only one memory region (or negative)? Ignore it */ | |
420 | if (nr_map < 2) | |
421 | return -1; | |
422 | ||
dc8e8120 | 423 | return __append_e820_map(biosmap, nr_map); |
b79cd8f1 YL |
424 | } |
425 | ||
773e673d | 426 | static u64 __init __e820_update_range(struct e820map *e820x, u64 start, |
fc9036ea YL |
427 | u64 size, unsigned old_type, |
428 | unsigned new_type) | |
b79cd8f1 | 429 | { |
78a8b35b | 430 | u64 end; |
773e673d | 431 | unsigned int i; |
b79cd8f1 YL |
432 | u64 real_updated_size = 0; |
433 | ||
434 | BUG_ON(old_type == new_type); | |
435 | ||
232b957a YL |
436 | if (size > (ULLONG_MAX - start)) |
437 | size = ULLONG_MAX - start; | |
438 | ||
78a8b35b | 439 | end = start + size; |
91eb0f67 BH |
440 | printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", |
441 | (unsigned long long) start, (unsigned long long) (end - 1)); | |
c61cf4cf YL |
442 | e820_print_type(old_type); |
443 | printk(KERN_CONT " ==> "); | |
444 | e820_print_type(new_type); | |
445 | printk(KERN_CONT "\n"); | |
446 | ||
5c0e6f03 | 447 | for (i = 0; i < e820x->nr_map; i++) { |
fc9036ea | 448 | struct e820entry *ei = &e820x->map[i]; |
b79cd8f1 | 449 | u64 final_start, final_end; |
78a8b35b YL |
450 | u64 ei_end; |
451 | ||
b79cd8f1 YL |
452 | if (ei->type != old_type) |
453 | continue; | |
78a8b35b YL |
454 | |
455 | ei_end = ei->addr + ei->size; | |
456 | /* totally covered by new range? */ | |
457 | if (ei->addr >= start && ei_end <= end) { | |
b79cd8f1 YL |
458 | ei->type = new_type; |
459 | real_updated_size += ei->size; | |
460 | continue; | |
461 | } | |
78a8b35b YL |
462 | |
463 | /* new range is totally covered? */ | |
464 | if (ei->addr < start && ei_end > end) { | |
465 | __e820_add_region(e820x, start, size, new_type); | |
466 | __e820_add_region(e820x, end, ei_end - end, ei->type); | |
467 | ei->size = start - ei->addr; | |
468 | real_updated_size += size; | |
469 | continue; | |
470 | } | |
471 | ||
b79cd8f1 YL |
472 | /* partially covered */ |
473 | final_start = max(start, ei->addr); | |
78a8b35b | 474 | final_end = min(end, ei_end); |
b79cd8f1 YL |
475 | if (final_start >= final_end) |
476 | continue; | |
5c0e6f03 | 477 | |
773e673d YL |
478 | __e820_add_region(e820x, final_start, final_end - final_start, |
479 | new_type); | |
5c0e6f03 | 480 | |
b79cd8f1 | 481 | real_updated_size += final_end - final_start; |
976dd4dc | 482 | |
773e673d YL |
483 | /* |
484 | * left range could be head or tail, so need to update | |
485 | * size at first. | |
486 | */ | |
487 | ei->size -= final_end - final_start; | |
976dd4dc YL |
488 | if (ei->addr < final_start) |
489 | continue; | |
490 | ei->addr = final_end; | |
b79cd8f1 YL |
491 | } |
492 | return real_updated_size; | |
493 | } | |
494 | ||
fc9036ea YL |
495 | u64 __init e820_update_range(u64 start, u64 size, unsigned old_type, |
496 | unsigned new_type) | |
497 | { | |
47533968 | 498 | return __e820_update_range(e820, start, size, old_type, new_type); |
fc9036ea YL |
499 | } |
500 | ||
501 | static u64 __init e820_update_range_saved(u64 start, u64 size, | |
502 | unsigned old_type, unsigned new_type) | |
503 | { | |
47533968 | 504 | return __e820_update_range(e820_saved, start, size, old_type, |
fc9036ea YL |
505 | new_type); |
506 | } | |
507 | ||
7a1fd986 YL |
508 | /* make e820 not cover the range */ |
509 | u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type, | |
510 | int checktype) | |
511 | { | |
512 | int i; | |
1b5576e6 | 513 | u64 end; |
7a1fd986 YL |
514 | u64 real_removed_size = 0; |
515 | ||
232b957a YL |
516 | if (size > (ULLONG_MAX - start)) |
517 | size = ULLONG_MAX - start; | |
518 | ||
1b5576e6 | 519 | end = start + size; |
91eb0f67 BH |
520 | printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", |
521 | (unsigned long long) start, (unsigned long long) (end - 1)); | |
9f3a5f52 YL |
522 | if (checktype) |
523 | e820_print_type(old_type); | |
1b5576e6 YL |
524 | printk(KERN_CONT "\n"); |
525 | ||
47533968 DV |
526 | for (i = 0; i < e820->nr_map; i++) { |
527 | struct e820entry *ei = &e820->map[i]; | |
7a1fd986 | 528 | u64 final_start, final_end; |
9f3a5f52 | 529 | u64 ei_end; |
7a1fd986 YL |
530 | |
531 | if (checktype && ei->type != old_type) | |
532 | continue; | |
9f3a5f52 YL |
533 | |
534 | ei_end = ei->addr + ei->size; | |
7a1fd986 | 535 | /* totally covered? */ |
9f3a5f52 | 536 | if (ei->addr >= start && ei_end <= end) { |
7a1fd986 YL |
537 | real_removed_size += ei->size; |
538 | memset(ei, 0, sizeof(struct e820entry)); | |
539 | continue; | |
540 | } | |
9f3a5f52 YL |
541 | |
542 | /* new range is totally covered? */ | |
543 | if (ei->addr < start && ei_end > end) { | |
544 | e820_add_region(end, ei_end - end, ei->type); | |
545 | ei->size = start - ei->addr; | |
546 | real_removed_size += size; | |
547 | continue; | |
548 | } | |
549 | ||
7a1fd986 YL |
550 | /* partially covered */ |
551 | final_start = max(start, ei->addr); | |
9f3a5f52 | 552 | final_end = min(end, ei_end); |
7a1fd986 YL |
553 | if (final_start >= final_end) |
554 | continue; | |
555 | real_removed_size += final_end - final_start; | |
556 | ||
9f3a5f52 YL |
557 | /* |
558 | * left range could be head or tail, so need to update | |
559 | * size at first. | |
560 | */ | |
7a1fd986 YL |
561 | ei->size -= final_end - final_start; |
562 | if (ei->addr < final_start) | |
563 | continue; | |
564 | ei->addr = final_end; | |
565 | } | |
566 | return real_removed_size; | |
567 | } | |
568 | ||
b79cd8f1 YL |
569 | void __init update_e820(void) |
570 | { | |
47533968 | 571 | if (sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map)) |
b79cd8f1 | 572 | return; |
91eb0f67 | 573 | printk(KERN_INFO "e820: modified physical RAM map:\n"); |
b79cd8f1 YL |
574 | e820_print_map("modified"); |
575 | } | |
fc9036ea YL |
576 | static void __init update_e820_saved(void) |
577 | { | |
47533968 DV |
578 | sanitize_e820_map(e820_saved->map, ARRAY_SIZE(e820_saved->map), |
579 | &e820_saved->nr_map); | |
fc9036ea | 580 | } |
fd6493e1 | 581 | #define MAX_GAP_END 0x100000000ull |
b79cd8f1 | 582 | /* |
fd6493e1 | 583 | * Search for a gap in the e820 memory space from start_addr to end_addr. |
b79cd8f1 | 584 | */ |
3381959d | 585 | __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize, |
fd6493e1 | 586 | unsigned long start_addr, unsigned long long end_addr) |
b79cd8f1 | 587 | { |
fd6493e1 | 588 | unsigned long long last; |
47533968 | 589 | int i = e820->nr_map; |
b79cd8f1 YL |
590 | int found = 0; |
591 | ||
fd6493e1 AK |
592 | last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END; |
593 | ||
b79cd8f1 | 594 | while (--i >= 0) { |
47533968 DV |
595 | unsigned long long start = e820->map[i].addr; |
596 | unsigned long long end = start + e820->map[i].size; | |
b79cd8f1 | 597 | |
3381959d AK |
598 | if (end < start_addr) |
599 | continue; | |
600 | ||
b79cd8f1 YL |
601 | /* |
602 | * Since "last" is at most 4GB, we know we'll | |
603 | * fit in 32 bits if this condition is true | |
604 | */ | |
605 | if (last > end) { | |
606 | unsigned long gap = last - end; | |
607 | ||
3381959d AK |
608 | if (gap >= *gapsize) { |
609 | *gapsize = gap; | |
610 | *gapstart = end; | |
b79cd8f1 YL |
611 | found = 1; |
612 | } | |
613 | } | |
614 | if (start < last) | |
615 | last = start; | |
616 | } | |
3381959d AK |
617 | return found; |
618 | } | |
619 | ||
620 | /* | |
621 | * Search for the biggest gap in the low 32 bits of the e820 | |
622 | * memory space. We pass this space to PCI to assign MMIO resources | |
623 | * for hotplug or unconfigured devices in. | |
624 | * Hopefully the BIOS let enough space left. | |
625 | */ | |
626 | __init void e820_setup_gap(void) | |
627 | { | |
5d423ccd | 628 | unsigned long gapstart, gapsize; |
3381959d AK |
629 | int found; |
630 | ||
631 | gapstart = 0x10000000; | |
632 | gapsize = 0x400000; | |
fd6493e1 | 633 | found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END); |
b79cd8f1 YL |
634 | |
635 | #ifdef CONFIG_X86_64 | |
636 | if (!found) { | |
c987d12f | 637 | gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024; |
ad361c98 | 638 | printk(KERN_ERR |
91eb0f67 BH |
639 | "e820: cannot find a gap in the 32bit address range\n" |
640 | "e820: PCI devices with unassigned 32bit BARs may break!\n"); | |
b79cd8f1 YL |
641 | } |
642 | #endif | |
643 | ||
644 | /* | |
5d423ccd | 645 | * e820_reserve_resources_late protect stolen RAM already |
b79cd8f1 | 646 | */ |
5d423ccd | 647 | pci_mem_start = gapstart; |
b79cd8f1 YL |
648 | |
649 | printk(KERN_INFO | |
91eb0f67 BH |
650 | "e820: [mem %#010lx-%#010lx] available for PCI devices\n", |
651 | gapstart, gapstart + gapsize - 1); | |
b79cd8f1 YL |
652 | } |
653 | ||
47533968 DV |
654 | /* |
655 | * Called late during init, in free_initmem(). | |
656 | * | |
657 | * Initial e820 and e820_saved are largish __initdata arrays. | |
658 | * Copy them to (usually much smaller) dynamically allocated area. | |
659 | * This is done after all tweaks we ever do to them: | |
660 | * all functions which modify them are __init functions, | |
661 | * they won't exist after this point. | |
662 | */ | |
663 | __init void e820_reallocate_tables(void) | |
664 | { | |
665 | struct e820map *n; | |
666 | int size; | |
667 | ||
668 | size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820->nr_map; | |
669 | n = kmalloc(size, GFP_KERNEL); | |
670 | BUG_ON(!n); | |
671 | memcpy(n, e820, size); | |
672 | e820 = n; | |
673 | ||
674 | size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820_saved->nr_map; | |
675 | n = kmalloc(size, GFP_KERNEL); | |
676 | BUG_ON(!n); | |
677 | memcpy(n, e820_saved, size); | |
678 | e820_saved = n; | |
679 | } | |
680 | ||
8c5beb50 HY |
681 | /** |
682 | * Because of the size limitation of struct boot_params, only first | |
683 | * 128 E820 memory entries are passed to kernel via | |
684 | * boot_params.e820_map, others are passed via SETUP_E820_EXT node of | |
685 | * linked list of struct setup_data, which is parsed here. | |
686 | */ | |
30e46b57 | 687 | void __init parse_e820_ext(u64 phys_addr, u32 data_len) |
8c5beb50 | 688 | { |
8c5beb50 HY |
689 | int entries; |
690 | struct e820entry *extmap; | |
30e46b57 | 691 | struct setup_data *sdata; |
8c5beb50 | 692 | |
30e46b57 | 693 | sdata = early_memremap(phys_addr, data_len); |
8c5beb50 | 694 | entries = sdata->len / sizeof(struct e820entry); |
8c5beb50 | 695 | extmap = (struct e820entry *)(sdata->data); |
dc8e8120 | 696 | __append_e820_map(extmap, entries); |
47533968 | 697 | sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map); |
8d4a40bc | 698 | early_memunmap(sdata, data_len); |
91eb0f67 | 699 | printk(KERN_INFO "e820: extended physical RAM map:\n"); |
8c5beb50 HY |
700 | e820_print_map("extended"); |
701 | } | |
702 | ||
bf62f398 YL |
703 | #if defined(CONFIG_X86_64) || \ |
704 | (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION)) | |
705 | /** | |
706 | * Find the ranges of physical addresses that do not correspond to | |
707 | * e820 RAM areas and mark the corresponding pages as nosave for | |
708 | * hibernation (32 bit) or software suspend and suspend to RAM (64 bit). | |
709 | * | |
710 | * This function requires the e820 map to be sorted and without any | |
84779575 | 711 | * overlapping entries. |
bf62f398 YL |
712 | */ |
713 | void __init e820_mark_nosave_regions(unsigned long limit_pfn) | |
714 | { | |
715 | int i; | |
84779575 | 716 | unsigned long pfn = 0; |
bf62f398 | 717 | |
47533968 DV |
718 | for (i = 0; i < e820->nr_map; i++) { |
719 | struct e820entry *ei = &e820->map[i]; | |
bf62f398 YL |
720 | |
721 | if (pfn < PFN_UP(ei->addr)) | |
722 | register_nosave_region(pfn, PFN_UP(ei->addr)); | |
723 | ||
724 | pfn = PFN_DOWN(ei->addr + ei->size); | |
ec776ef6 | 725 | |
28bb2237 | 726 | if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN) |
bf62f398 YL |
727 | register_nosave_region(PFN_UP(ei->addr), pfn); |
728 | ||
729 | if (pfn >= limit_pfn) | |
730 | break; | |
731 | } | |
732 | } | |
733 | #endif | |
a4c81cf6 | 734 | |
b54ac6d2 | 735 | #ifdef CONFIG_ACPI |
b69edc76 RW |
736 | /** |
737 | * Mark ACPI NVS memory region, so that we can save/restore it during | |
738 | * hibernation and the subsequent resume. | |
739 | */ | |
740 | static int __init e820_mark_nvs_memory(void) | |
741 | { | |
742 | int i; | |
743 | ||
47533968 DV |
744 | for (i = 0; i < e820->nr_map; i++) { |
745 | struct e820entry *ei = &e820->map[i]; | |
b69edc76 RW |
746 | |
747 | if (ei->type == E820_NVS) | |
b54ac6d2 | 748 | acpi_nvs_register(ei->addr, ei->size); |
b69edc76 RW |
749 | } |
750 | ||
751 | return 0; | |
752 | } | |
753 | core_initcall(e820_mark_nvs_memory); | |
754 | #endif | |
755 | ||
2944e16b | 756 | /* |
72d7c3b3 | 757 | * pre allocated 4k and reserved it in memblock and e820_saved |
2944e16b | 758 | */ |
ab5d140b | 759 | u64 __init early_reserve_e820(u64 size, u64 align) |
2944e16b | 760 | { |
2944e16b | 761 | u64 addr; |
2944e16b | 762 | |
ab5d140b TH |
763 | addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
764 | if (addr) { | |
765 | e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED); | |
91eb0f67 | 766 | printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n"); |
ab5d140b | 767 | update_e820_saved(); |
61438766 | 768 | } |
2944e16b | 769 | |
2944e16b YL |
770 | return addr; |
771 | } | |
772 | ||
ee0c80fa YL |
773 | #ifdef CONFIG_X86_32 |
774 | # ifdef CONFIG_X86_PAE | |
775 | # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT)) | |
776 | # else | |
777 | # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT)) | |
778 | # endif | |
779 | #else /* CONFIG_X86_32 */ | |
bd70e522 | 780 | # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT |
ee0c80fa YL |
781 | #endif |
782 | ||
ee0c80fa YL |
783 | /* |
784 | * Find the highest page frame number we have available | |
785 | */ | |
917db484 | 786 | static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type) |
ee0c80fa | 787 | { |
2dc807b3 YL |
788 | int i; |
789 | unsigned long last_pfn = 0; | |
ee0c80fa YL |
790 | unsigned long max_arch_pfn = MAX_ARCH_PFN; |
791 | ||
47533968 DV |
792 | for (i = 0; i < e820->nr_map; i++) { |
793 | struct e820entry *ei = &e820->map[i]; | |
f361a450 | 794 | unsigned long start_pfn; |
2dc807b3 YL |
795 | unsigned long end_pfn; |
796 | ||
917db484 | 797 | if (ei->type != type) |
c22d4c18 | 798 | continue; |
c22d4c18 | 799 | |
f361a450 | 800 | start_pfn = ei->addr >> PAGE_SHIFT; |
2dc807b3 | 801 | end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT; |
f361a450 YL |
802 | |
803 | if (start_pfn >= limit_pfn) | |
804 | continue; | |
805 | if (end_pfn > limit_pfn) { | |
806 | last_pfn = limit_pfn; | |
807 | break; | |
808 | } | |
2dc807b3 YL |
809 | if (end_pfn > last_pfn) |
810 | last_pfn = end_pfn; | |
811 | } | |
ee0c80fa YL |
812 | |
813 | if (last_pfn > max_arch_pfn) | |
814 | last_pfn = max_arch_pfn; | |
ee0c80fa | 815 | |
91eb0f67 | 816 | printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n", |
ee0c80fa YL |
817 | last_pfn, max_arch_pfn); |
818 | return last_pfn; | |
819 | } | |
f361a450 YL |
820 | unsigned long __init e820_end_of_ram_pfn(void) |
821 | { | |
917db484 | 822 | return e820_end_pfn(MAX_ARCH_PFN, E820_RAM); |
f361a450 | 823 | } |
ee0c80fa | 824 | |
f361a450 YL |
825 | unsigned long __init e820_end_of_low_ram_pfn(void) |
826 | { | |
917db484 | 827 | return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_RAM); |
f361a450 | 828 | } |
ee0c80fa | 829 | |
8c2103f2 | 830 | static void __init early_panic(char *msg) |
ab4a465e YL |
831 | { |
832 | early_printk(msg); | |
833 | panic(msg); | |
834 | } | |
835 | ||
69a7704d YL |
836 | static int userdef __initdata; |
837 | ||
ab4a465e YL |
838 | /* "mem=nopentium" disables the 4MB page tables. */ |
839 | static int __init parse_memopt(char *p) | |
840 | { | |
841 | u64 mem_size; | |
842 | ||
843 | if (!p) | |
844 | return -EINVAL; | |
845 | ||
ab4a465e | 846 | if (!strcmp(p, "nopentium")) { |
9a6d44b9 | 847 | #ifdef CONFIG_X86_32 |
ab4a465e YL |
848 | setup_clear_cpu_cap(X86_FEATURE_PSE); |
849 | return 0; | |
9a6d44b9 KM |
850 | #else |
851 | printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n"); | |
852 | return -EINVAL; | |
ab4a465e | 853 | #endif |
9a6d44b9 | 854 | } |
ab4a465e | 855 | |
69a7704d | 856 | userdef = 1; |
ab4a465e | 857 | mem_size = memparse(p, &p); |
77eed821 KM |
858 | /* don't remove all of memory when handling "mem={invalid}" param */ |
859 | if (mem_size == 0) | |
860 | return -EINVAL; | |
69a7704d | 861 | e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1); |
611dfd78 | 862 | |
ab4a465e YL |
863 | return 0; |
864 | } | |
865 | early_param("mem", parse_memopt); | |
866 | ||
9710f581 | 867 | static int __init parse_memmap_one(char *p) |
ab4a465e YL |
868 | { |
869 | char *oldp; | |
870 | u64 start_at, mem_size; | |
871 | ||
a737abd1 CG |
872 | if (!p) |
873 | return -EINVAL; | |
874 | ||
d6be118a | 875 | if (!strncmp(p, "exactmap", 8)) { |
ab4a465e YL |
876 | #ifdef CONFIG_CRASH_DUMP |
877 | /* | |
878 | * If we are doing a crash dump, we still need to know | |
879 | * the real mem size before original memory map is | |
880 | * reset. | |
881 | */ | |
f361a450 | 882 | saved_max_pfn = e820_end_of_ram_pfn(); |
ab4a465e | 883 | #endif |
47533968 | 884 | e820->nr_map = 0; |
ab4a465e YL |
885 | userdef = 1; |
886 | return 0; | |
887 | } | |
888 | ||
889 | oldp = p; | |
890 | mem_size = memparse(p, &p); | |
891 | if (p == oldp) | |
892 | return -EINVAL; | |
893 | ||
894 | userdef = 1; | |
895 | if (*p == '@') { | |
896 | start_at = memparse(p+1, &p); | |
d0be6bde | 897 | e820_add_region(start_at, mem_size, E820_RAM); |
ab4a465e YL |
898 | } else if (*p == '#') { |
899 | start_at = memparse(p+1, &p); | |
d0be6bde | 900 | e820_add_region(start_at, mem_size, E820_ACPI); |
ab4a465e YL |
901 | } else if (*p == '$') { |
902 | start_at = memparse(p+1, &p); | |
d0be6bde | 903 | e820_add_region(start_at, mem_size, E820_RESERVED); |
ec776ef6 CH |
904 | } else if (*p == '!') { |
905 | start_at = memparse(p+1, &p); | |
906 | e820_add_region(start_at, mem_size, E820_PRAM); | |
7b479bec | 907 | } else |
69a7704d | 908 | e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1); |
7b479bec | 909 | |
ab4a465e YL |
910 | return *p == '\0' ? 0 : -EINVAL; |
911 | } | |
9710f581 YL |
912 | static int __init parse_memmap_opt(char *str) |
913 | { | |
914 | while (str) { | |
915 | char *k = strchr(str, ','); | |
916 | ||
917 | if (k) | |
918 | *k++ = 0; | |
919 | ||
920 | parse_memmap_one(str); | |
921 | str = k; | |
922 | } | |
923 | ||
924 | return 0; | |
925 | } | |
ab4a465e YL |
926 | early_param("memmap", parse_memmap_opt); |
927 | ||
928 | void __init finish_e820_parsing(void) | |
929 | { | |
930 | if (userdef) { | |
47533968 DV |
931 | if (sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), |
932 | &e820->nr_map) < 0) | |
ab4a465e | 933 | early_panic("Invalid user supplied memory map"); |
ab4a465e | 934 | |
91eb0f67 | 935 | printk(KERN_INFO "e820: user-defined physical RAM map:\n"); |
ab4a465e YL |
936 | e820_print_map("user"); |
937 | } | |
938 | } | |
41c094fd | 939 | |
8c2103f2 | 940 | static const char *__init e820_type_to_string(int e820_type) |
5dfcf14d BW |
941 | { |
942 | switch (e820_type) { | |
943 | case E820_RESERVED_KERN: | |
944 | case E820_RAM: return "System RAM"; | |
945 | case E820_ACPI: return "ACPI Tables"; | |
946 | case E820_NVS: return "ACPI Non-volatile Storage"; | |
671eef85 | 947 | case E820_UNUSABLE: return "Unusable memory"; |
ad5fb870 DW |
948 | case E820_PRAM: return "Persistent Memory (legacy)"; |
949 | case E820_PMEM: return "Persistent Memory"; | |
5dfcf14d BW |
950 | default: return "reserved"; |
951 | } | |
952 | } | |
953 | ||
8c2103f2 | 954 | static unsigned long __init e820_type_to_iomem_type(int e820_type) |
f33b14a4 TK |
955 | { |
956 | switch (e820_type) { | |
957 | case E820_RESERVED_KERN: | |
958 | case E820_RAM: | |
959 | return IORESOURCE_SYSTEM_RAM; | |
960 | case E820_ACPI: | |
961 | case E820_NVS: | |
962 | case E820_UNUSABLE: | |
963 | case E820_PRAM: | |
964 | case E820_PMEM: | |
965 | default: | |
966 | return IORESOURCE_MEM; | |
967 | } | |
968 | } | |
969 | ||
8c2103f2 | 970 | static unsigned long __init e820_type_to_iores_desc(int e820_type) |
f33b14a4 TK |
971 | { |
972 | switch (e820_type) { | |
973 | case E820_ACPI: | |
974 | return IORES_DESC_ACPI_TABLES; | |
975 | case E820_NVS: | |
976 | return IORES_DESC_ACPI_NV_STORAGE; | |
977 | case E820_PMEM: | |
978 | return IORES_DESC_PERSISTENT_MEMORY; | |
979 | case E820_PRAM: | |
980 | return IORES_DESC_PERSISTENT_MEMORY_LEGACY; | |
981 | case E820_RESERVED_KERN: | |
982 | case E820_RAM: | |
983 | case E820_UNUSABLE: | |
984 | default: | |
985 | return IORES_DESC_NONE; | |
986 | } | |
987 | } | |
988 | ||
8c2103f2 | 989 | static bool __init do_mark_busy(u32 type, struct resource *res) |
ad5fb870 DW |
990 | { |
991 | /* this is the legacy bios/dos rom-shadow + mmio region */ | |
992 | if (res->start < (1ULL<<20)) | |
993 | return true; | |
994 | ||
995 | /* | |
996 | * Treat persistent memory like device memory, i.e. reserve it | |
997 | * for exclusive use of a driver | |
998 | */ | |
999 | switch (type) { | |
1000 | case E820_RESERVED: | |
1001 | case E820_PRAM: | |
1002 | case E820_PMEM: | |
1003 | return false; | |
1004 | default: | |
1005 | return true; | |
1006 | } | |
1007 | } | |
1008 | ||
41c094fd YL |
1009 | /* |
1010 | * Mark e820 reserved areas as busy for the resource manager. | |
1011 | */ | |
a5444d15 | 1012 | static struct resource __initdata *e820_res; |
41c094fd YL |
1013 | void __init e820_reserve_resources(void) |
1014 | { | |
1015 | int i; | |
58f7c988 | 1016 | struct resource *res; |
a5444d15 | 1017 | u64 end; |
41c094fd | 1018 | |
47533968 | 1019 | res = alloc_bootmem(sizeof(struct resource) * e820->nr_map); |
58f7c988 | 1020 | e820_res = res; |
47533968 DV |
1021 | for (i = 0; i < e820->nr_map; i++) { |
1022 | end = e820->map[i].addr + e820->map[i].size - 1; | |
8308c54d | 1023 | if (end != (resource_size_t)end) { |
41c094fd YL |
1024 | res++; |
1025 | continue; | |
1026 | } | |
47533968 DV |
1027 | res->name = e820_type_to_string(e820->map[i].type); |
1028 | res->start = e820->map[i].addr; | |
b4df32f4 YL |
1029 | res->end = end; |
1030 | ||
47533968 DV |
1031 | res->flags = e820_type_to_iomem_type(e820->map[i].type); |
1032 | res->desc = e820_type_to_iores_desc(e820->map[i].type); | |
a5444d15 IM |
1033 | |
1034 | /* | |
1035 | * don't register the region that could be conflicted with | |
1036 | * pci device BAR resource and insert them later in | |
1037 | * pcibios_resource_survey() | |
1038 | */ | |
47533968 | 1039 | if (do_mark_busy(e820->map[i].type, res)) { |
1f987577 | 1040 | res->flags |= IORESOURCE_BUSY; |
58f7c988 | 1041 | insert_resource(&iomem_resource, res); |
1f987577 | 1042 | } |
41c094fd YL |
1043 | res++; |
1044 | } | |
5dfcf14d | 1045 | |
47533968 DV |
1046 | for (i = 0; i < e820_saved->nr_map; i++) { |
1047 | struct e820entry *entry = &e820_saved->map[i]; | |
5dfcf14d | 1048 | firmware_map_add_early(entry->addr, |
4ed940d4 | 1049 | entry->addr + entry->size, |
5dfcf14d BW |
1050 | e820_type_to_string(entry->type)); |
1051 | } | |
41c094fd YL |
1052 | } |
1053 | ||
45fbe3ee | 1054 | /* How much should we pad RAM ending depending on where it is? */ |
8c2103f2 | 1055 | static unsigned long __init ram_alignment(resource_size_t pos) |
45fbe3ee LT |
1056 | { |
1057 | unsigned long mb = pos >> 20; | |
1058 | ||
1059 | /* To 64kB in the first megabyte */ | |
1060 | if (!mb) | |
1061 | return 64*1024; | |
1062 | ||
1063 | /* To 1MB in the first 16MB */ | |
1064 | if (mb < 16) | |
1065 | return 1024*1024; | |
1066 | ||
15b812f1 YL |
1067 | /* To 64MB for anything above that */ |
1068 | return 64*1024*1024; | |
45fbe3ee LT |
1069 | } |
1070 | ||
7c5371c4 YL |
1071 | #define MAX_RESOURCE_SIZE ((resource_size_t)-1) |
1072 | ||
58f7c988 YL |
1073 | void __init e820_reserve_resources_late(void) |
1074 | { | |
1075 | int i; | |
1076 | struct resource *res; | |
1077 | ||
1078 | res = e820_res; | |
47533968 | 1079 | for (i = 0; i < e820->nr_map; i++) { |
a5444d15 | 1080 | if (!res->parent && res->end) |
1f987577 | 1081 | insert_resource_expand_to_fit(&iomem_resource, res); |
58f7c988 YL |
1082 | res++; |
1083 | } | |
45fbe3ee LT |
1084 | |
1085 | /* | |
1086 | * Try to bump up RAM regions to reasonable boundaries to | |
1087 | * avoid stolen RAM: | |
1088 | */ | |
47533968 DV |
1089 | for (i = 0; i < e820->nr_map; i++) { |
1090 | struct e820entry *entry = &e820->map[i]; | |
7c5371c4 | 1091 | u64 start, end; |
45fbe3ee LT |
1092 | |
1093 | if (entry->type != E820_RAM) | |
1094 | continue; | |
1095 | start = entry->addr + entry->size; | |
7c5371c4 YL |
1096 | end = round_up(start, ram_alignment(start)) - 1; |
1097 | if (end > MAX_RESOURCE_SIZE) | |
1098 | end = MAX_RESOURCE_SIZE; | |
1099 | if (start >= end) | |
45fbe3ee | 1100 | continue; |
91eb0f67 BH |
1101 | printk(KERN_DEBUG |
1102 | "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", | |
1103 | start, end); | |
7c5371c4 YL |
1104 | reserve_region_with_split(&iomem_resource, start, end, |
1105 | "RAM buffer"); | |
45fbe3ee | 1106 | } |
58f7c988 YL |
1107 | } |
1108 | ||
95a71a45 | 1109 | char *__init default_machine_specific_memory_setup(void) |
064d25f1 YL |
1110 | { |
1111 | char *who = "BIOS-e820"; | |
ba639039 | 1112 | u32 new_nr; |
064d25f1 YL |
1113 | /* |
1114 | * Try to copy the BIOS-supplied E820-map. | |
1115 | * | |
1116 | * Otherwise fake a memory map; one section from 0k->640k, | |
1117 | * the next section from 1mb->appropriate_mem_k | |
1118 | */ | |
1119 | new_nr = boot_params.e820_entries; | |
1120 | sanitize_e820_map(boot_params.e820_map, | |
1121 | ARRAY_SIZE(boot_params.e820_map), | |
1122 | &new_nr); | |
1123 | boot_params.e820_entries = new_nr; | |
dc8e8120 YL |
1124 | if (append_e820_map(boot_params.e820_map, boot_params.e820_entries) |
1125 | < 0) { | |
95a71a45 | 1126 | u64 mem_size; |
064d25f1 YL |
1127 | |
1128 | /* compare results from other methods and take the greater */ | |
1129 | if (boot_params.alt_mem_k | |
1130 | < boot_params.screen_info.ext_mem_k) { | |
1131 | mem_size = boot_params.screen_info.ext_mem_k; | |
1132 | who = "BIOS-88"; | |
1133 | } else { | |
1134 | mem_size = boot_params.alt_mem_k; | |
1135 | who = "BIOS-e801"; | |
1136 | } | |
1137 | ||
47533968 | 1138 | e820->nr_map = 0; |
064d25f1 YL |
1139 | e820_add_region(0, LOWMEMSIZE(), E820_RAM); |
1140 | e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM); | |
064d25f1 YL |
1141 | } |
1142 | ||
1143 | /* In case someone cares... */ | |
1144 | return who; | |
1145 | } | |
1146 | ||
064d25f1 YL |
1147 | void __init setup_memory_map(void) |
1148 | { | |
0be15526 YL |
1149 | char *who; |
1150 | ||
6b18ae3e | 1151 | who = x86_init.resources.memory_setup(); |
47533968 | 1152 | memcpy(e820_saved, e820, sizeof(struct e820map)); |
91eb0f67 | 1153 | printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n"); |
0be15526 | 1154 | e820_print_map(who); |
064d25f1 | 1155 | } |
72d7c3b3 YL |
1156 | |
1157 | void __init memblock_x86_fill(void) | |
1158 | { | |
1159 | int i; | |
1160 | u64 end; | |
1161 | ||
1162 | /* | |
1163 | * EFI may have more than 128 entries | |
1164 | * We are safe to enable resizing, beause memblock_x86_fill() | |
1165 | * is rather later for x86 | |
1166 | */ | |
1aadc056 | 1167 | memblock_allow_resize(); |
72d7c3b3 | 1168 | |
47533968 DV |
1169 | for (i = 0; i < e820->nr_map; i++) { |
1170 | struct e820entry *ei = &e820->map[i]; | |
72d7c3b3 YL |
1171 | |
1172 | end = ei->addr + ei->size; | |
1173 | if (end != (resource_size_t)end) | |
1174 | continue; | |
1175 | ||
1176 | if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN) | |
1177 | continue; | |
1178 | ||
1179 | memblock_add(ei->addr, ei->size); | |
1180 | } | |
1181 | ||
6ede1fd3 YL |
1182 | /* throw away partial pages */ |
1183 | memblock_trim_memory(PAGE_SIZE); | |
1184 | ||
72d7c3b3 YL |
1185 | memblock_dump_all(); |
1186 | } | |
6f2a7536 YL |
1187 | |
1188 | void __init memblock_find_dma_reserve(void) | |
1189 | { | |
1190 | #ifdef CONFIG_X86_64 | |
6b5d41a1 TH |
1191 | u64 nr_pages = 0, nr_free_pages = 0; |
1192 | unsigned long start_pfn, end_pfn; | |
1193 | phys_addr_t start, end; | |
1194 | int i; | |
1195 | u64 u; | |
1196 | ||
6f2a7536 YL |
1197 | /* |
1198 | * need to find out used area below MAX_DMA_PFN | |
1199 | * need to use memblock to get free size in [0, MAX_DMA_PFN] | |
1200 | * at first, and assume boot_mem will not take below MAX_DMA_PFN | |
1201 | */ | |
6b5d41a1 | 1202 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { |
29258cf4 XQ |
1203 | start_pfn = min(start_pfn, MAX_DMA_PFN); |
1204 | end_pfn = min(end_pfn, MAX_DMA_PFN); | |
6b5d41a1 TH |
1205 | nr_pages += end_pfn - start_pfn; |
1206 | } | |
1207 | ||
fc6daaf9 TL |
1208 | for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, |
1209 | NULL) { | |
6b5d41a1 TH |
1210 | start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN); |
1211 | end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN); | |
1212 | if (start_pfn < end_pfn) | |
1213 | nr_free_pages += end_pfn - start_pfn; | |
1214 | } | |
1215 | ||
1216 | set_dma_reserve(nr_pages - nr_free_pages); | |
6f2a7536 YL |
1217 | #endif |
1218 | } |