Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Handle the memory map. | |
3 | * The functions here do the job until bootmem takes over. | |
4 | * $Id: e820.c,v 1.4 2002/09/19 19:25:32 ak Exp $ | |
5 | */ | |
6 | #include <linux/config.h> | |
7 | #include <linux/kernel.h> | |
8 | #include <linux/types.h> | |
9 | #include <linux/init.h> | |
10 | #include <linux/bootmem.h> | |
11 | #include <linux/ioport.h> | |
12 | #include <linux/string.h> | |
13 | #include <asm/page.h> | |
14 | #include <asm/e820.h> | |
15 | #include <asm/proto.h> | |
16 | #include <asm/bootsetup.h> | |
17 | ||
18 | extern char _end[]; | |
19 | ||
20 | /* | |
21 | * PFN of last memory page. | |
22 | */ | |
23 | unsigned long end_pfn; | |
24 | ||
25 | /* | |
26 | * end_pfn only includes RAM, while end_pfn_map includes all e820 entries. | |
27 | * The direct mapping extends to end_pfn_map, so that we can directly access | |
28 | * apertures, ACPI and other tables without having to play with fixmaps. | |
29 | */ | |
30 | unsigned long end_pfn_map; | |
31 | ||
32 | /* | |
33 | * Last pfn which the user wants to use. | |
34 | */ | |
35 | unsigned long end_user_pfn = MAXMEM>>PAGE_SHIFT; | |
36 | ||
37 | extern struct resource code_resource, data_resource; | |
38 | ||
39 | /* Check for some hardcoded bad areas that early boot is not allowed to touch */ | |
40 | static inline int bad_addr(unsigned long *addrp, unsigned long size) | |
41 | { | |
42 | unsigned long addr = *addrp, last = addr + size; | |
43 | ||
44 | /* various gunk below that needed for SMP startup */ | |
45 | if (addr < 0x8000) { | |
46 | *addrp = 0x8000; | |
47 | return 1; | |
48 | } | |
49 | ||
50 | /* direct mapping tables of the kernel */ | |
51 | if (last >= table_start<<PAGE_SHIFT && addr < table_end<<PAGE_SHIFT) { | |
52 | *addrp = table_end << PAGE_SHIFT; | |
53 | return 1; | |
54 | } | |
55 | ||
56 | /* initrd */ | |
57 | #ifdef CONFIG_BLK_DEV_INITRD | |
58 | if (LOADER_TYPE && INITRD_START && last >= INITRD_START && | |
59 | addr < INITRD_START+INITRD_SIZE) { | |
60 | *addrp = INITRD_START + INITRD_SIZE; | |
61 | return 1; | |
62 | } | |
63 | #endif | |
64 | /* kernel code + 640k memory hole (later should not be needed, but | |
65 | be paranoid for now) */ | |
66 | if (last >= 640*1024 && addr < __pa_symbol(&_end)) { | |
67 | *addrp = __pa_symbol(&_end); | |
68 | return 1; | |
69 | } | |
70 | /* XXX ramdisk image here? */ | |
71 | return 0; | |
72 | } | |
73 | ||
74 | int __init e820_mapped(unsigned long start, unsigned long end, unsigned type) | |
75 | { | |
76 | int i; | |
77 | for (i = 0; i < e820.nr_map; i++) { | |
78 | struct e820entry *ei = &e820.map[i]; | |
79 | if (type && ei->type != type) | |
80 | continue; | |
81 | if (ei->addr >= end || ei->addr + ei->size < start) | |
82 | continue; | |
83 | return 1; | |
84 | } | |
85 | return 0; | |
86 | } | |
87 | ||
88 | /* | |
89 | * Find a free area in a specific range. | |
90 | */ | |
91 | unsigned long __init find_e820_area(unsigned long start, unsigned long end, unsigned size) | |
92 | { | |
93 | int i; | |
94 | for (i = 0; i < e820.nr_map; i++) { | |
95 | struct e820entry *ei = &e820.map[i]; | |
96 | unsigned long addr = ei->addr, last; | |
97 | if (ei->type != E820_RAM) | |
98 | continue; | |
99 | if (addr < start) | |
100 | addr = start; | |
101 | if (addr > ei->addr + ei->size) | |
102 | continue; | |
103 | while (bad_addr(&addr, size) && addr+size < ei->addr + ei->size) | |
104 | ; | |
105 | last = addr + size; | |
106 | if (last > ei->addr + ei->size) | |
107 | continue; | |
108 | if (last > end) | |
109 | continue; | |
110 | return addr; | |
111 | } | |
112 | return -1UL; | |
113 | } | |
114 | ||
115 | /* | |
116 | * Free bootmem based on the e820 table for a node. | |
117 | */ | |
118 | void __init e820_bootmem_free(pg_data_t *pgdat, unsigned long start,unsigned long end) | |
119 | { | |
120 | int i; | |
121 | for (i = 0; i < e820.nr_map; i++) { | |
122 | struct e820entry *ei = &e820.map[i]; | |
123 | unsigned long last, addr; | |
124 | ||
125 | if (ei->type != E820_RAM || | |
126 | ei->addr+ei->size <= start || | |
127 | ei->addr > end) | |
128 | continue; | |
129 | ||
130 | addr = round_up(ei->addr, PAGE_SIZE); | |
131 | if (addr < start) | |
132 | addr = start; | |
133 | ||
134 | last = round_down(ei->addr + ei->size, PAGE_SIZE); | |
135 | if (last >= end) | |
136 | last = end; | |
137 | ||
138 | if (last > addr && last-addr >= PAGE_SIZE) | |
139 | free_bootmem_node(pgdat, addr, last-addr); | |
140 | } | |
141 | } | |
142 | ||
143 | /* | |
144 | * Find the highest page frame number we have available | |
145 | */ | |
146 | unsigned long __init e820_end_of_ram(void) | |
147 | { | |
148 | int i; | |
149 | unsigned long end_pfn = 0; | |
150 | ||
151 | for (i = 0; i < e820.nr_map; i++) { | |
152 | struct e820entry *ei = &e820.map[i]; | |
153 | unsigned long start, end; | |
154 | ||
155 | start = round_up(ei->addr, PAGE_SIZE); | |
156 | end = round_down(ei->addr + ei->size, PAGE_SIZE); | |
157 | if (start >= end) | |
158 | continue; | |
159 | if (ei->type == E820_RAM) { | |
160 | if (end > end_pfn<<PAGE_SHIFT) | |
161 | end_pfn = end>>PAGE_SHIFT; | |
162 | } else { | |
163 | if (end > end_pfn_map<<PAGE_SHIFT) | |
164 | end_pfn_map = end>>PAGE_SHIFT; | |
165 | } | |
166 | } | |
167 | ||
168 | if (end_pfn > end_pfn_map) | |
169 | end_pfn_map = end_pfn; | |
170 | if (end_pfn_map > MAXMEM>>PAGE_SHIFT) | |
171 | end_pfn_map = MAXMEM>>PAGE_SHIFT; | |
172 | if (end_pfn > end_user_pfn) | |
173 | end_pfn = end_user_pfn; | |
174 | if (end_pfn > end_pfn_map) | |
175 | end_pfn = end_pfn_map; | |
176 | ||
177 | return end_pfn; | |
178 | } | |
179 | ||
180 | /* | |
181 | * Mark e820 reserved areas as busy for the resource manager. | |
182 | */ | |
183 | void __init e820_reserve_resources(void) | |
184 | { | |
185 | int i; | |
186 | for (i = 0; i < e820.nr_map; i++) { | |
187 | struct resource *res; | |
188 | if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL) | |
189 | continue; | |
190 | res = alloc_bootmem_low(sizeof(struct resource)); | |
191 | switch (e820.map[i].type) { | |
192 | case E820_RAM: res->name = "System RAM"; break; | |
193 | case E820_ACPI: res->name = "ACPI Tables"; break; | |
194 | case E820_NVS: res->name = "ACPI Non-volatile Storage"; break; | |
195 | default: res->name = "reserved"; | |
196 | } | |
197 | res->start = e820.map[i].addr; | |
198 | res->end = res->start + e820.map[i].size - 1; | |
199 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; | |
200 | request_resource(&iomem_resource, res); | |
201 | if (e820.map[i].type == E820_RAM) { | |
202 | /* | |
203 | * We don't know which RAM region contains kernel data, | |
204 | * so we try it repeatedly and let the resource manager | |
205 | * test it. | |
206 | */ | |
207 | request_resource(res, &code_resource); | |
208 | request_resource(res, &data_resource); | |
209 | } | |
210 | } | |
211 | } | |
212 | ||
213 | /* | |
214 | * Add a memory region to the kernel e820 map. | |
215 | */ | |
216 | void __init add_memory_region(unsigned long start, unsigned long size, int type) | |
217 | { | |
218 | int x = e820.nr_map; | |
219 | ||
220 | if (x == E820MAX) { | |
221 | printk(KERN_ERR "Ooops! Too many entries in the memory map!\n"); | |
222 | return; | |
223 | } | |
224 | ||
225 | e820.map[x].addr = start; | |
226 | e820.map[x].size = size; | |
227 | e820.map[x].type = type; | |
228 | e820.nr_map++; | |
229 | } | |
230 | ||
231 | void __init e820_print_map(char *who) | |
232 | { | |
233 | int i; | |
234 | ||
235 | for (i = 0; i < e820.nr_map; i++) { | |
236 | printk(" %s: %016Lx - %016Lx ", who, | |
237 | (unsigned long long) e820.map[i].addr, | |
238 | (unsigned long long) (e820.map[i].addr + e820.map[i].size)); | |
239 | switch (e820.map[i].type) { | |
240 | case E820_RAM: printk("(usable)\n"); | |
241 | break; | |
242 | case E820_RESERVED: | |
243 | printk("(reserved)\n"); | |
244 | break; | |
245 | case E820_ACPI: | |
246 | printk("(ACPI data)\n"); | |
247 | break; | |
248 | case E820_NVS: | |
249 | printk("(ACPI NVS)\n"); | |
250 | break; | |
251 | default: printk("type %u\n", e820.map[i].type); | |
252 | break; | |
253 | } | |
254 | } | |
255 | } | |
256 | ||
257 | /* | |
258 | * Sanitize the BIOS e820 map. | |
259 | * | |
260 | * Some e820 responses include overlapping entries. The following | |
261 | * replaces the original e820 map with a new one, removing overlaps. | |
262 | * | |
263 | */ | |
264 | static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map) | |
265 | { | |
266 | struct change_member { | |
267 | struct e820entry *pbios; /* pointer to original bios entry */ | |
268 | unsigned long long addr; /* address for this change point */ | |
269 | }; | |
270 | static struct change_member change_point_list[2*E820MAX] __initdata; | |
271 | static struct change_member *change_point[2*E820MAX] __initdata; | |
272 | static struct e820entry *overlap_list[E820MAX] __initdata; | |
273 | static struct e820entry new_bios[E820MAX] __initdata; | |
274 | struct change_member *change_tmp; | |
275 | unsigned long current_type, last_type; | |
276 | unsigned long long last_addr; | |
277 | int chgidx, still_changing; | |
278 | int overlap_entries; | |
279 | int new_bios_entry; | |
280 | int old_nr, new_nr; | |
281 | int i; | |
282 | ||
283 | /* | |
284 | Visually we're performing the following (1,2,3,4 = memory types)... | |
285 | ||
286 | Sample memory map (w/overlaps): | |
287 | ____22__________________ | |
288 | ______________________4_ | |
289 | ____1111________________ | |
290 | _44_____________________ | |
291 | 11111111________________ | |
292 | ____________________33__ | |
293 | ___________44___________ | |
294 | __________33333_________ | |
295 | ______________22________ | |
296 | ___________________2222_ | |
297 | _________111111111______ | |
298 | _____________________11_ | |
299 | _________________4______ | |
300 | ||
301 | Sanitized equivalent (no overlap): | |
302 | 1_______________________ | |
303 | _44_____________________ | |
304 | ___1____________________ | |
305 | ____22__________________ | |
306 | ______11________________ | |
307 | _________1______________ | |
308 | __________3_____________ | |
309 | ___________44___________ | |
310 | _____________33_________ | |
311 | _______________2________ | |
312 | ________________1_______ | |
313 | _________________4______ | |
314 | ___________________2____ | |
315 | ____________________33__ | |
316 | ______________________4_ | |
317 | */ | |
318 | ||
319 | /* if there's only one memory region, don't bother */ | |
320 | if (*pnr_map < 2) | |
321 | return -1; | |
322 | ||
323 | old_nr = *pnr_map; | |
324 | ||
325 | /* bail out if we find any unreasonable addresses in bios map */ | |
326 | for (i=0; i<old_nr; i++) | |
327 | if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) | |
328 | return -1; | |
329 | ||
330 | /* create pointers for initial change-point information (for sorting) */ | |
331 | for (i=0; i < 2*old_nr; i++) | |
332 | change_point[i] = &change_point_list[i]; | |
333 | ||
334 | /* record all known change-points (starting and ending addresses) */ | |
335 | chgidx = 0; | |
336 | for (i=0; i < old_nr; i++) { | |
337 | change_point[chgidx]->addr = biosmap[i].addr; | |
338 | change_point[chgidx++]->pbios = &biosmap[i]; | |
339 | change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size; | |
340 | change_point[chgidx++]->pbios = &biosmap[i]; | |
341 | } | |
342 | ||
343 | /* sort change-point list by memory addresses (low -> high) */ | |
344 | still_changing = 1; | |
345 | while (still_changing) { | |
346 | still_changing = 0; | |
347 | for (i=1; i < 2*old_nr; i++) { | |
348 | /* if <current_addr> > <last_addr>, swap */ | |
349 | /* or, if current=<start_addr> & last=<end_addr>, swap */ | |
350 | if ((change_point[i]->addr < change_point[i-1]->addr) || | |
351 | ((change_point[i]->addr == change_point[i-1]->addr) && | |
352 | (change_point[i]->addr == change_point[i]->pbios->addr) && | |
353 | (change_point[i-1]->addr != change_point[i-1]->pbios->addr)) | |
354 | ) | |
355 | { | |
356 | change_tmp = change_point[i]; | |
357 | change_point[i] = change_point[i-1]; | |
358 | change_point[i-1] = change_tmp; | |
359 | still_changing=1; | |
360 | } | |
361 | } | |
362 | } | |
363 | ||
364 | /* create a new bios memory map, removing overlaps */ | |
365 | overlap_entries=0; /* number of entries in the overlap table */ | |
366 | new_bios_entry=0; /* index for creating new bios map entries */ | |
367 | last_type = 0; /* start with undefined memory type */ | |
368 | last_addr = 0; /* start with 0 as last starting address */ | |
369 | /* loop through change-points, determining affect on the new bios map */ | |
370 | for (chgidx=0; chgidx < 2*old_nr; chgidx++) | |
371 | { | |
372 | /* keep track of all overlapping bios entries */ | |
373 | if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr) | |
374 | { | |
375 | /* add map entry to overlap list (> 1 entry implies an overlap) */ | |
376 | overlap_list[overlap_entries++]=change_point[chgidx]->pbios; | |
377 | } | |
378 | else | |
379 | { | |
380 | /* remove entry from list (order independent, so swap with last) */ | |
381 | for (i=0; i<overlap_entries; i++) | |
382 | { | |
383 | if (overlap_list[i] == change_point[chgidx]->pbios) | |
384 | overlap_list[i] = overlap_list[overlap_entries-1]; | |
385 | } | |
386 | overlap_entries--; | |
387 | } | |
388 | /* if there are overlapping entries, decide which "type" to use */ | |
389 | /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */ | |
390 | current_type = 0; | |
391 | for (i=0; i<overlap_entries; i++) | |
392 | if (overlap_list[i]->type > current_type) | |
393 | current_type = overlap_list[i]->type; | |
394 | /* continue building up new bios map based on this information */ | |
395 | if (current_type != last_type) { | |
396 | if (last_type != 0) { | |
397 | new_bios[new_bios_entry].size = | |
398 | change_point[chgidx]->addr - last_addr; | |
399 | /* move forward only if the new size was non-zero */ | |
400 | if (new_bios[new_bios_entry].size != 0) | |
401 | if (++new_bios_entry >= E820MAX) | |
402 | break; /* no more space left for new bios entries */ | |
403 | } | |
404 | if (current_type != 0) { | |
405 | new_bios[new_bios_entry].addr = change_point[chgidx]->addr; | |
406 | new_bios[new_bios_entry].type = current_type; | |
407 | last_addr=change_point[chgidx]->addr; | |
408 | } | |
409 | last_type = current_type; | |
410 | } | |
411 | } | |
412 | new_nr = new_bios_entry; /* retain count for new bios entries */ | |
413 | ||
414 | /* copy new bios mapping into original location */ | |
415 | memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry)); | |
416 | *pnr_map = new_nr; | |
417 | ||
418 | return 0; | |
419 | } | |
420 | ||
421 | /* | |
422 | * Copy the BIOS e820 map into a safe place. | |
423 | * | |
424 | * Sanity-check it while we're at it.. | |
425 | * | |
426 | * If we're lucky and live on a modern system, the setup code | |
427 | * will have given us a memory map that we can use to properly | |
428 | * set up memory. If we aren't, we'll fake a memory map. | |
429 | * | |
430 | * We check to see that the memory map contains at least 2 elements | |
431 | * before we'll use it, because the detection code in setup.S may | |
432 | * not be perfect and most every PC known to man has two memory | |
433 | * regions: one from 0 to 640k, and one from 1mb up. (The IBM | |
434 | * thinkpad 560x, for example, does not cooperate with the memory | |
435 | * detection code.) | |
436 | */ | |
437 | static int __init copy_e820_map(struct e820entry * biosmap, int nr_map) | |
438 | { | |
439 | /* Only one memory region (or negative)? Ignore it */ | |
440 | if (nr_map < 2) | |
441 | return -1; | |
442 | ||
443 | do { | |
444 | unsigned long start = biosmap->addr; | |
445 | unsigned long size = biosmap->size; | |
446 | unsigned long end = start + size; | |
447 | unsigned long type = biosmap->type; | |
448 | ||
449 | /* Overflow in 64 bits? Ignore the memory map. */ | |
450 | if (start > end) | |
451 | return -1; | |
452 | ||
453 | /* | |
454 | * Some BIOSes claim RAM in the 640k - 1M region. | |
455 | * Not right. Fix it up. | |
456 | * | |
457 | * This should be removed on Hammer which is supposed to not | |
458 | * have non e820 covered ISA mappings there, but I had some strange | |
459 | * problems so it stays for now. -AK | |
460 | */ | |
461 | if (type == E820_RAM) { | |
462 | if (start < 0x100000ULL && end > 0xA0000ULL) { | |
463 | if (start < 0xA0000ULL) | |
464 | add_memory_region(start, 0xA0000ULL-start, type); | |
465 | if (end <= 0x100000ULL) | |
466 | continue; | |
467 | start = 0x100000ULL; | |
468 | size = end - start; | |
469 | } | |
470 | } | |
471 | ||
472 | add_memory_region(start, size, type); | |
473 | } while (biosmap++,--nr_map); | |
474 | return 0; | |
475 | } | |
476 | ||
477 | void __init setup_memory_region(void) | |
478 | { | |
479 | char *who = "BIOS-e820"; | |
480 | ||
481 | /* | |
482 | * Try to copy the BIOS-supplied E820-map. | |
483 | * | |
484 | * Otherwise fake a memory map; one section from 0k->640k, | |
485 | * the next section from 1mb->appropriate_mem_k | |
486 | */ | |
487 | sanitize_e820_map(E820_MAP, &E820_MAP_NR); | |
488 | if (copy_e820_map(E820_MAP, E820_MAP_NR) < 0) { | |
489 | unsigned long mem_size; | |
490 | ||
491 | /* compare results from other methods and take the greater */ | |
492 | if (ALT_MEM_K < EXT_MEM_K) { | |
493 | mem_size = EXT_MEM_K; | |
494 | who = "BIOS-88"; | |
495 | } else { | |
496 | mem_size = ALT_MEM_K; | |
497 | who = "BIOS-e801"; | |
498 | } | |
499 | ||
500 | e820.nr_map = 0; | |
501 | add_memory_region(0, LOWMEMSIZE(), E820_RAM); | |
502 | add_memory_region(HIGH_MEMORY, mem_size << 10, E820_RAM); | |
503 | } | |
504 | printk(KERN_INFO "BIOS-provided physical RAM map:\n"); | |
505 | e820_print_map(who); | |
506 | } | |
507 | ||
508 | void __init parse_memopt(char *p, char **from) | |
509 | { | |
510 | end_user_pfn = memparse(p, from); | |
511 | end_user_pfn >>= PAGE_SHIFT; | |
512 | } | |
513 |