~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/kernel/e820.c

Version: ~ [ linux-5.5-rc7 ] ~ [ linux-5.4.13 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.97 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.166 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.210 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.210 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.81 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Low level x86 E820 memory map handling functions.
  3  *
  4  * The firmware and bootloader passes us the "E820 table", which is the primary
  5  * physical memory layout description available about x86 systems.
  6  *
  7  * The kernel takes the E820 memory layout and optionally modifies it with
  8  * quirks and other tweaks, and feeds that into the generic Linux memory
  9  * allocation code routines via a platform independent interface (memblock, etc.).
 10  */
 11 #include <linux/crash_dump.h>
 12 #include <linux/memblock.h>
 13 #include <linux/suspend.h>
 14 #include <linux/acpi.h>
 15 #include <linux/firmware-map.h>
 16 #include <linux/sort.h>
 17 
 18 #include <asm/e820/api.h>
 19 #include <asm/setup.h>
 20 
 21 /*
 22  * We organize the E820 table into three main data structures:
 23  *
 24  * - 'e820_table_firmware': the original firmware version passed to us by the
 25  *   bootloader - not modified by the kernel. It is composed of two parts:
 26  *   the first 128 E820 memory entries in boot_params.e820_table and the remaining
 27  *   (if any) entries of the SETUP_E820_EXT nodes. We use this to:
 28  *
 29  *       - inform the user about the firmware's notion of memory layout
 30  *         via /sys/firmware/memmap
 31  *
 32  *       - the hibernation code uses it to generate a kernel-independent MD5
 33  *         fingerprint of the physical memory layout of a system.
 34  *
 35  * - 'e820_table_kexec': a slightly modified (by the kernel) firmware version
 36  *   passed to us by the bootloader - the major difference between
 37  *   e820_table_firmware[] and this one is that, the latter marks the setup_data
 38  *   list created by the EFI boot stub as reserved, so that kexec can reuse the
 39  *   setup_data information in the second kernel. Besides, e820_table_kexec[]
 40  *   might also be modified by the kexec itself to fake a mptable.
 41  *   We use this to:
 42  *
 43  *       - kexec, which is a bootloader in disguise, uses the original E820
 44  *         layout to pass to the kexec-ed kernel. This way the original kernel
 45  *         can have a restricted E820 map while the kexec()-ed kexec-kernel
 46  *         can have access to full memory - etc.
 47  *
 48  * - 'e820_table': this is the main E820 table that is massaged by the
 49  *   low level x86 platform code, or modified by boot parameters, before
 50  *   passed on to higher level MM layers.
 51  *
 52  * Once the E820 map has been converted to the standard Linux memory layout
 53  * information its role stops - modifying it has no effect and does not get
 54  * re-propagated. So itsmain role is a temporary bootstrap storage of firmware
 55  * specific memory layout data during early bootup.
 56  */
 57 static struct e820_table e820_table_init                __initdata;
 58 static struct e820_table e820_table_kexec_init          __initdata;
 59 static struct e820_table e820_table_firmware_init       __initdata;
 60 
 61 struct e820_table *e820_table __refdata                 = &e820_table_init;
 62 struct e820_table *e820_table_kexec __refdata           = &e820_table_kexec_init;
 63 struct e820_table *e820_table_firmware __refdata        = &e820_table_firmware_init;
 64 
 65 /* For PCI or other memory-mapped resources */
 66 unsigned long pci_mem_start = 0xaeedbabe;
 67 #ifdef CONFIG_PCI
 68 EXPORT_SYMBOL(pci_mem_start);
 69 #endif
 70 
 71 /*
 72  * This function checks if any part of the range <start,end> is mapped
 73  * with type.
 74  */
 75 bool e820__mapped_any(u64 start, u64 end, enum e820_type type)
 76 {
 77         int i;
 78 
 79         for (i = 0; i < e820_table->nr_entries; i++) {
 80                 struct e820_entry *entry = &e820_table->entries[i];
 81 
 82                 if (type && entry->type != type)
 83                         continue;
 84                 if (entry->addr >= end || entry->addr + entry->size <= start)
 85                         continue;
 86                 return 1;
 87         }
 88         return 0;
 89 }
 90 EXPORT_SYMBOL_GPL(e820__mapped_any);
 91 
 92 /*
 93  * This function checks if the entire <start,end> range is mapped with 'type'.
 94  *
 95  * Note: this function only works correctly once the E820 table is sorted and
 96  * not-overlapping (at least for the range specified), which is the case normally.
 97  */
 98 static struct e820_entry *__e820__mapped_all(u64 start, u64 end,
 99                                              enum e820_type type)
100 {
101         int i;
102 
103         for (i = 0; i < e820_table->nr_entries; i++) {
104                 struct e820_entry *entry = &e820_table->entries[i];
105 
106                 if (type && entry->type != type)
107                         continue;
108 
109                 /* Is the region (part) in overlap with the current region? */
110                 if (entry->addr >= end || entry->addr + entry->size <= start)
111                         continue;
112 
113                 /*
114                  * If the region is at the beginning of <start,end> we move
115                  * 'start' to the end of the region since it's ok until there
116                  */
117                 if (entry->addr <= start)
118                         start = entry->addr + entry->size;
119 
120                 /*
121                  * If 'start' is now at or beyond 'end', we're done, full
122                  * coverage of the desired range exists:
123                  */
124                 if (start >= end)
125                         return entry;
126         }
127 
128         return NULL;
129 }
130 
131 /*
132  * This function checks if the entire range <start,end> is mapped with type.
133  */
134 bool __init e820__mapped_all(u64 start, u64 end, enum e820_type type)
135 {
136         return __e820__mapped_all(start, end, type);
137 }
138 
139 /*
140  * This function returns the type associated with the range <start,end>.
141  */
142 int e820__get_entry_type(u64 start, u64 end)
143 {
144         struct e820_entry *entry = __e820__mapped_all(start, end, 0);
145 
146         return entry ? entry->type : -EINVAL;
147 }
148 
149 /*
150  * Add a memory region to the kernel E820 map.
151  */
152 static void __init __e820__range_add(struct e820_table *table, u64 start, u64 size, enum e820_type type)
153 {
154         int x = table->nr_entries;
155 
156         if (x >= ARRAY_SIZE(table->entries)) {
157                 pr_err("too many entries; ignoring [mem %#010llx-%#010llx]\n",
158                        start, start + size - 1);
159                 return;
160         }
161 
162         table->entries[x].addr = start;
163         table->entries[x].size = size;
164         table->entries[x].type = type;
165         table->nr_entries++;
166 }
167 
168 void __init e820__range_add(u64 start, u64 size, enum e820_type type)
169 {
170         __e820__range_add(e820_table, start, size, type);
171 }
172 
173 static void __init e820_print_type(enum e820_type type)
174 {
175         switch (type) {
176         case E820_TYPE_RAM:             /* Fall through: */
177         case E820_TYPE_RESERVED_KERN:   pr_cont("usable");                      break;
178         case E820_TYPE_RESERVED:        pr_cont("reserved");                    break;
179         case E820_TYPE_ACPI:            pr_cont("ACPI data");                   break;
180         case E820_TYPE_NVS:             pr_cont("ACPI NVS");                    break;
181         case E820_TYPE_UNUSABLE:        pr_cont("unusable");                    break;
182         case E820_TYPE_PMEM:            /* Fall through: */
183         case E820_TYPE_PRAM:            pr_cont("persistent (type %u)", type);  break;
184         default:                        pr_cont("type %u", type);               break;
185         }
186 }
187 
188 void __init e820__print_table(char *who)
189 {
190         int i;
191 
192         for (i = 0; i < e820_table->nr_entries; i++) {
193                 pr_info("%s: [mem %#018Lx-%#018Lx] ",
194                         who,
195                         e820_table->entries[i].addr,
196                         e820_table->entries[i].addr + e820_table->entries[i].size - 1);
197 
198                 e820_print_type(e820_table->entries[i].type);
199                 pr_cont("\n");
200         }
201 }
202 
203 /*
204  * Sanitize an E820 map.
205  *
206  * Some E820 layouts include overlapping entries. The following
207  * replaces the original E820 map with a new one, removing overlaps,
208  * and resolving conflicting memory types in favor of highest
209  * numbered type.
210  *
211  * The input parameter 'entries' points to an array of 'struct
212  * e820_entry' which on entry has elements in the range [0, *nr_entries)
213  * valid, and which has space for up to max_nr_entries entries.
214  * On return, the resulting sanitized E820 map entries will be in
215  * overwritten in the same location, starting at 'entries'.
216  *
217  * The integer pointed to by nr_entries must be valid on entry (the
218  * current number of valid entries located at 'entries'). If the
219  * sanitizing succeeds the *nr_entries will be updated with the new
220  * number of valid entries (something no more than max_nr_entries).
221  *
222  * The return value from e820__update_table() is zero if it
223  * successfully 'sanitized' the map entries passed in, and is -1
224  * if it did nothing, which can happen if either of (1) it was
225  * only passed one map entry, or (2) any of the input map entries
226  * were invalid (start + size < start, meaning that the size was
227  * so big the described memory range wrapped around through zero.)
228  *
229  *      Visually we're performing the following
230  *      (1,2,3,4 = memory types)...
231  *
232  *      Sample memory map (w/overlaps):
233  *         ____22__________________
234  *         ______________________4_
235  *         ____1111________________
236  *         _44_____________________
237  *         11111111________________
238  *         ____________________33__
239  *         ___________44___________
240  *         __________33333_________
241  *         ______________22________
242  *         ___________________2222_
243  *         _________111111111______
244  *         _____________________11_
245  *         _________________4______
246  *
247  *      Sanitized equivalent (no overlap):
248  *         1_______________________
249  *         _44_____________________
250  *         ___1____________________
251  *         ____22__________________
252  *         ______11________________
253  *         _________1______________
254  *         __________3_____________
255  *         ___________44___________
256  *         _____________33_________
257  *         _______________2________
258  *         ________________1_______
259  *         _________________4______
260  *         ___________________2____
261  *         ____________________33__
262  *         ______________________4_
263  */
264 struct change_member {
265         /* Pointer to the original entry: */
266         struct e820_entry       *entry;
267         /* Address for this change point: */
268         unsigned long long      addr;
269 };
270 
271 static struct change_member     change_point_list[2*E820_MAX_ENTRIES]   __initdata;
272 static struct change_member     *change_point[2*E820_MAX_ENTRIES]       __initdata;
273 static struct e820_entry        *overlap_list[E820_MAX_ENTRIES]         __initdata;
274 static struct e820_entry        new_entries[E820_MAX_ENTRIES]           __initdata;
275 
276 static int __init cpcompare(const void *a, const void *b)
277 {
278         struct change_member * const *app = a, * const *bpp = b;
279         const struct change_member *ap = *app, *bp = *bpp;
280 
281         /*
282          * Inputs are pointers to two elements of change_point[].  If their
283          * addresses are not equal, their difference dominates.  If the addresses
284          * are equal, then consider one that represents the end of its region
285          * to be greater than one that does not.
286          */
287         if (ap->addr != bp->addr)
288                 return ap->addr > bp->addr ? 1 : -1;
289 
290         return (ap->addr != ap->entry->addr) - (bp->addr != bp->entry->addr);
291 }
292 
293 int __init e820__update_table(struct e820_table *table)
294 {
295         struct e820_entry *entries = table->entries;
296         u32 max_nr_entries = ARRAY_SIZE(table->entries);
297         enum e820_type current_type, last_type;
298         unsigned long long last_addr;
299         u32 new_nr_entries, overlap_entries;
300         u32 i, chg_idx, chg_nr;
301 
302         /* If there's only one memory region, don't bother: */
303         if (table->nr_entries < 2)
304                 return -1;
305 
306         BUG_ON(table->nr_entries > max_nr_entries);
307 
308         /* Bail out if we find any unreasonable addresses in the map: */
309         for (i = 0; i < table->nr_entries; i++) {
310                 if (entries[i].addr + entries[i].size < entries[i].addr)
311                         return -1;
312         }
313 
314         /* Create pointers for initial change-point information (for sorting): */
315         for (i = 0; i < 2 * table->nr_entries; i++)
316                 change_point[i] = &change_point_list[i];
317 
318         /*
319          * Record all known change-points (starting and ending addresses),
320          * omitting empty memory regions:
321          */
322         chg_idx = 0;
323         for (i = 0; i < table->nr_entries; i++) {
324                 if (entries[i].size != 0) {
325                         change_point[chg_idx]->addr     = entries[i].addr;
326                         change_point[chg_idx++]->entry  = &entries[i];
327                         change_point[chg_idx]->addr     = entries[i].addr + entries[i].size;
328                         change_point[chg_idx++]->entry  = &entries[i];
329                 }
330         }
331         chg_nr = chg_idx;
332 
333         /* Sort change-point list by memory addresses (low -> high): */
334         sort(change_point, chg_nr, sizeof(*change_point), cpcompare, NULL);
335 
336         /* Create a new memory map, removing overlaps: */
337         overlap_entries = 0;     /* Number of entries in the overlap table */
338         new_nr_entries = 0;      /* Index for creating new map entries */
339         last_type = 0;           /* Start with undefined memory type */
340         last_addr = 0;           /* Start with 0 as last starting address */
341 
342         /* Loop through change-points, determining effect on the new map: */
343         for (chg_idx = 0; chg_idx < chg_nr; chg_idx++) {
344                 /* Keep track of all overlapping entries */
345                 if (change_point[chg_idx]->addr == change_point[chg_idx]->entry->addr) {
346                         /* Add map entry to overlap list (> 1 entry implies an overlap) */
347                         overlap_list[overlap_entries++] = change_point[chg_idx]->entry;
348                 } else {
349                         /* Remove entry from list (order independent, so swap with last): */
350                         for (i = 0; i < overlap_entries; i++) {
351                                 if (overlap_list[i] == change_point[chg_idx]->entry)
352                                         overlap_list[i] = overlap_list[overlap_entries-1];
353                         }
354                         overlap_entries--;
355                 }
356                 /*
357                  * If there are overlapping entries, decide which
358                  * "type" to use (larger value takes precedence --
359                  * 1=usable, 2,3,4,4+=unusable)
360                  */
361                 current_type = 0;
362                 for (i = 0; i < overlap_entries; i++) {
363                         if (overlap_list[i]->type > current_type)
364                                 current_type = overlap_list[i]->type;
365                 }
366 
367                 /* Continue building up new map based on this information: */
368                 if (current_type != last_type || current_type == E820_TYPE_PRAM) {
369                         if (last_type != 0)      {
370                                 new_entries[new_nr_entries].size = change_point[chg_idx]->addr - last_addr;
371                                 /* Move forward only if the new size was non-zero: */
372                                 if (new_entries[new_nr_entries].size != 0)
373                                         /* No more space left for new entries? */
374                                         if (++new_nr_entries >= max_nr_entries)
375                                                 break;
376                         }
377                         if (current_type != 0)  {
378                                 new_entries[new_nr_entries].addr = change_point[chg_idx]->addr;
379                                 new_entries[new_nr_entries].type = current_type;
380                                 last_addr = change_point[chg_idx]->addr;
381                         }
382                         last_type = current_type;
383                 }
384         }
385 
386         /* Copy the new entries into the original location: */
387         memcpy(entries, new_entries, new_nr_entries*sizeof(*entries));
388         table->nr_entries = new_nr_entries;
389 
390         return 0;
391 }
392 
393 static int __init __append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
394 {
395         struct boot_e820_entry *entry = entries;
396 
397         while (nr_entries) {
398                 u64 start = entry->addr;
399                 u64 size = entry->size;
400                 u64 end = start + size - 1;
401                 u32 type = entry->type;
402 
403                 /* Ignore the entry on 64-bit overflow: */
404                 if (start > end && likely(size))
405                         return -1;
406 
407                 e820__range_add(start, size, type);
408 
409                 entry++;
410                 nr_entries--;
411         }
412         return 0;
413 }
414 
415 /*
416  * Copy the BIOS E820 map into a safe place.
417  *
418  * Sanity-check it while we're at it..
419  *
420  * If we're lucky and live on a modern system, the setup code
421  * will have given us a memory map that we can use to properly
422  * set up memory.  If we aren't, we'll fake a memory map.
423  */
424 static int __init append_e820_table(struct boot_e820_entry *entries, u32 nr_entries)
425 {
426         /* Only one memory region (or negative)? Ignore it */
427         if (nr_entries < 2)
428                 return -1;
429 
430         return __append_e820_table(entries, nr_entries);
431 }
432 
433 static u64 __init
434 __e820__range_update(struct e820_table *table, u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
435 {
436         u64 end;
437         unsigned int i;
438         u64 real_updated_size = 0;
439 
440         BUG_ON(old_type == new_type);
441 
442         if (size > (ULLONG_MAX - start))
443                 size = ULLONG_MAX - start;
444 
445         end = start + size;
446         printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ", start, end - 1);
447         e820_print_type(old_type);
448         pr_cont(" ==> ");
449         e820_print_type(new_type);
450         pr_cont("\n");
451 
452         for (i = 0; i < table->nr_entries; i++) {
453                 struct e820_entry *entry = &table->entries[i];
454                 u64 final_start, final_end;
455                 u64 entry_end;
456 
457                 if (entry->type != old_type)
458                         continue;
459 
460                 entry_end = entry->addr + entry->size;
461 
462                 /* Completely covered by new range? */
463                 if (entry->addr >= start && entry_end <= end) {
464                         entry->type = new_type;
465                         real_updated_size += entry->size;
466                         continue;
467                 }
468 
469                 /* New range is completely covered? */
470                 if (entry->addr < start && entry_end > end) {
471                         __e820__range_add(table, start, size, new_type);
472                         __e820__range_add(table, end, entry_end - end, entry->type);
473                         entry->size = start - entry->addr;
474                         real_updated_size += size;
475                         continue;
476                 }
477 
478                 /* Partially covered: */
479                 final_start = max(start, entry->addr);
480                 final_end = min(end, entry_end);
481                 if (final_start >= final_end)
482                         continue;
483 
484                 __e820__range_add(table, final_start, final_end - final_start, new_type);
485 
486                 real_updated_size += final_end - final_start;
487 
488                 /*
489                  * Left range could be head or tail, so need to update
490                  * its size first:
491                  */
492                 entry->size -= final_end - final_start;
493                 if (entry->addr < final_start)
494                         continue;
495 
496                 entry->addr = final_end;
497         }
498         return real_updated_size;
499 }
500 
501 u64 __init e820__range_update(u64 start, u64 size, enum e820_type old_type, enum e820_type new_type)
502 {
503         return __e820__range_update(e820_table, start, size, old_type, new_type);
504 }
505 
506 static u64 __init e820__range_update_kexec(u64 start, u64 size, enum e820_type old_type, enum e820_type  new_type)
507 {
508         return __e820__range_update(e820_table_kexec, start, size, old_type, new_type);
509 }
510 
511 /* Remove a range of memory from the E820 table: */
512 u64 __init e820__range_remove(u64 start, u64 size, enum e820_type old_type, bool check_type)
513 {
514         int i;
515         u64 end;
516         u64 real_removed_size = 0;
517 
518         if (size > (ULLONG_MAX - start))
519                 size = ULLONG_MAX - start;
520 
521         end = start + size;
522         printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ", start, end - 1);
523         if (check_type)
524                 e820_print_type(old_type);
525         pr_cont("\n");
526 
527         for (i = 0; i < e820_table->nr_entries; i++) {
528                 struct e820_entry *entry = &e820_table->entries[i];
529                 u64 final_start, final_end;
530                 u64 entry_end;
531 
532                 if (check_type && entry->type != old_type)
533                         continue;
534 
535                 entry_end = entry->addr + entry->size;
536 
537                 /* Completely covered? */
538                 if (entry->addr >= start && entry_end <= end) {
539                         real_removed_size += entry->size;
540                         memset(entry, 0, sizeof(*entry));
541                         continue;
542                 }
543 
544                 /* Is the new range completely covered? */
545                 if (entry->addr < start && entry_end > end) {
546                         e820__range_add(end, entry_end - end, entry->type);
547                         entry->size = start - entry->addr;
548                         real_removed_size += size;
549                         continue;
550                 }
551 
552                 /* Partially covered: */
553                 final_start = max(start, entry->addr);
554                 final_end = min(end, entry_end);
555                 if (final_start >= final_end)
556                         continue;
557 
558                 real_removed_size += final_end - final_start;
559 
560                 /*
561                  * Left range could be head or tail, so need to update
562                  * the size first:
563                  */
564                 entry->size -= final_end - final_start;
565                 if (entry->addr < final_start)
566                         continue;
567 
568                 entry->addr = final_end;
569         }
570         return real_removed_size;
571 }
572 
573 void __init e820__update_table_print(void)
574 {
575         if (e820__update_table(e820_table))
576                 return;
577 
578         pr_info("modified physical RAM map:\n");
579         e820__print_table("modified");
580 }
581 
582 static void __init e820__update_table_kexec(void)
583 {
584         e820__update_table(e820_table_kexec);
585 }
586 
587 #define MAX_GAP_END 0x100000000ull
588 
589 /*
590  * Search for a gap in the E820 memory space from 0 to MAX_GAP_END (4GB).
591  */
592 static int __init e820_search_gap(unsigned long *gapstart, unsigned long *gapsize)
593 {
594         unsigned long long last = MAX_GAP_END;
595         int i = e820_table->nr_entries;
596         int found = 0;
597 
598         while (--i >= 0) {
599                 unsigned long long start = e820_table->entries[i].addr;
600                 unsigned long long end = start + e820_table->entries[i].size;
601 
602                 /*
603                  * Since "last" is at most 4GB, we know we'll
604                  * fit in 32 bits if this condition is true:
605                  */
606                 if (last > end) {
607                         unsigned long gap = last - end;
608 
609                         if (gap >= *gapsize) {
610                                 *gapsize = gap;
611                                 *gapstart = end;
612                                 found = 1;
613                         }
614                 }
615                 if (start < last)
616                         last = start;
617         }
618         return found;
619 }
620 
621 /*
622  * Search for the biggest gap in the low 32 bits of the E820
623  * memory space. We pass this space to the PCI subsystem, so
624  * that it can assign MMIO resources for hotplug or
625  * unconfigured devices in.
626  *
627  * Hopefully the BIOS let enough space left.
628  */
629 __init void e820__setup_pci_gap(void)
630 {
631         unsigned long gapstart, gapsize;
632         int found;
633 
634         gapsize = 0x400000;
635         found  = e820_search_gap(&gapstart, &gapsize);
636 
637         if (!found) {
638 #ifdef CONFIG_X86_64
639                 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
640                 pr_err("Cannot find an available gap in the 32-bit address range\n");
641                 pr_err("PCI devices with unassigned 32-bit BARs may not work!\n");
642 #else
643                 gapstart = 0x10000000;
644 #endif
645         }
646 
647         /*
648          * e820__reserve_resources_late() protects stolen RAM already:
649          */
650         pci_mem_start = gapstart;
651 
652         pr_info("[mem %#010lx-%#010lx] available for PCI devices\n",
653                 gapstart, gapstart + gapsize - 1);
654 }
655 
656 /*
657  * Called late during init, in free_initmem().
658  *
659  * Initial e820_table and e820_table_kexec are largish __initdata arrays.
660  *
661  * Copy them to a (usually much smaller) dynamically allocated area that is
662  * sized precisely after the number of e820 entries.
663  *
664  * This is done after we've performed all the fixes and tweaks to the tables.
665  * All functions which modify them are __init functions, which won't exist
666  * after free_initmem().
667  */
668 __init void e820__reallocate_tables(void)
669 {
670         struct e820_table *n;
671         int size;
672 
673         size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table->nr_entries;
674         n = kmalloc(size, GFP_KERNEL);
675         BUG_ON(!n);
676         memcpy(n, e820_table, size);
677         e820_table = n;
678 
679         size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_kexec->nr_entries;
680         n = kmalloc(size, GFP_KERNEL);
681         BUG_ON(!n);
682         memcpy(n, e820_table_kexec, size);
683         e820_table_kexec = n;
684 
685         size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry)*e820_table_firmware->nr_entries;
686         n = kmalloc(size, GFP_KERNEL);
687         BUG_ON(!n);
688         memcpy(n, e820_table_firmware, size);
689         e820_table_firmware = n;
690 }
691 
692 /*
693  * Because of the small fixed size of struct boot_params, only the first
694  * 128 E820 memory entries are passed to the kernel via boot_params.e820_table,
695  * the remaining (if any) entries are passed via the SETUP_E820_EXT node of
696  * struct setup_data, which is parsed here.
697  */
698 void __init e820__memory_setup_extended(u64 phys_addr, u32 data_len)
699 {
700         int entries;
701         struct boot_e820_entry *extmap;
702         struct setup_data *sdata;
703 
704         sdata = early_memremap(phys_addr, data_len);
705         entries = sdata->len / sizeof(*extmap);
706         extmap = (struct boot_e820_entry *)(sdata->data);
707 
708         __append_e820_table(extmap, entries);
709         e820__update_table(e820_table);
710 
711         memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
712         memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
713 
714         early_memunmap(sdata, data_len);
715         pr_info("extended physical RAM map:\n");
716         e820__print_table("extended");
717 }
718 
719 /*
720  * Find the ranges of physical addresses that do not correspond to
721  * E820 RAM areas and register the corresponding pages as 'nosave' for
722  * hibernation (32-bit) or software suspend and suspend to RAM (64-bit).
723  *
724  * This function requires the E820 map to be sorted and without any
725  * overlapping entries.
726  */
727 void __init e820__register_nosave_regions(unsigned long limit_pfn)
728 {
729         int i;
730         unsigned long pfn = 0;
731 
732         for (i = 0; i < e820_table->nr_entries; i++) {
733                 struct e820_entry *entry = &e820_table->entries[i];
734 
735                 if (pfn < PFN_UP(entry->addr))
736                         register_nosave_region(pfn, PFN_UP(entry->addr));
737 
738                 pfn = PFN_DOWN(entry->addr + entry->size);
739 
740                 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
741                         register_nosave_region(PFN_UP(entry->addr), pfn);
742 
743                 if (pfn >= limit_pfn)
744                         break;
745         }
746 }
747 
748 #ifdef CONFIG_ACPI
749 /*
750  * Register ACPI NVS memory regions, so that we can save/restore them during
751  * hibernation and the subsequent resume:
752  */
753 static int __init e820__register_nvs_regions(void)
754 {
755         int i;
756 
757         for (i = 0; i < e820_table->nr_entries; i++) {
758                 struct e820_entry *entry = &e820_table->entries[i];
759 
760                 if (entry->type == E820_TYPE_NVS)
761                         acpi_nvs_register(entry->addr, entry->size);
762         }
763 
764         return 0;
765 }
766 core_initcall(e820__register_nvs_regions);
767 #endif
768 
769 /*
770  * Allocate the requested number of bytes with the requsted alignment
771  * and return (the physical address) to the caller. Also register this
772  * range in the 'kexec' E820 table as a reserved range.
773  *
774  * This allows kexec to fake a new mptable, as if it came from the real
775  * system.
776  */
777 u64 __init e820__memblock_alloc_reserved(u64 size, u64 align)
778 {
779         u64 addr;
780 
781         addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
782         if (addr) {
783                 e820__range_update_kexec(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
784                 pr_info("update e820_table_kexec for e820__memblock_alloc_reserved()\n");
785                 e820__update_table_kexec();
786         }
787 
788         return addr;
789 }
790 
791 #ifdef CONFIG_X86_32
792 # ifdef CONFIG_X86_PAE
793 #  define MAX_ARCH_PFN          (1ULL<<(36-PAGE_SHIFT))
794 # else
795 #  define MAX_ARCH_PFN          (1ULL<<(32-PAGE_SHIFT))
796 # endif
797 #else /* CONFIG_X86_32 */
798 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
799 #endif
800 
801 /*
802  * Find the highest page frame number we have available
803  */
804 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, enum e820_type type)
805 {
806         int i;
807         unsigned long last_pfn = 0;
808         unsigned long max_arch_pfn = MAX_ARCH_PFN;
809 
810         for (i = 0; i < e820_table->nr_entries; i++) {
811                 struct e820_entry *entry = &e820_table->entries[i];
812                 unsigned long start_pfn;
813                 unsigned long end_pfn;
814 
815                 if (entry->type != type)
816                         continue;
817 
818                 start_pfn = entry->addr >> PAGE_SHIFT;
819                 end_pfn = (entry->addr + entry->size) >> PAGE_SHIFT;
820 
821                 if (start_pfn >= limit_pfn)
822                         continue;
823                 if (end_pfn > limit_pfn) {
824                         last_pfn = limit_pfn;
825                         break;
826                 }
827                 if (end_pfn > last_pfn)
828                         last_pfn = end_pfn;
829         }
830 
831         if (last_pfn > max_arch_pfn)
832                 last_pfn = max_arch_pfn;
833 
834         pr_info("last_pfn = %#lx max_arch_pfn = %#lx\n",
835                 last_pfn, max_arch_pfn);
836         return last_pfn;
837 }
838 
839 unsigned long __init e820__end_of_ram_pfn(void)
840 {
841         return e820_end_pfn(MAX_ARCH_PFN, E820_TYPE_RAM);
842 }
843 
844 unsigned long __init e820__end_of_low_ram_pfn(void)
845 {
846         return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_TYPE_RAM);
847 }
848 
849 static void __init early_panic(char *msg)
850 {
851         early_printk(msg);
852         panic(msg);
853 }
854 
855 static int userdef __initdata;
856 
857 /* The "mem=nopentium" boot option disables 4MB page tables on 32-bit kernels: */
858 static int __init parse_memopt(char *p)
859 {
860         u64 mem_size;
861 
862         if (!p)
863                 return -EINVAL;
864 
865         if (!strcmp(p, "nopentium")) {
866 #ifdef CONFIG_X86_32
867                 setup_clear_cpu_cap(X86_FEATURE_PSE);
868                 return 0;
869 #else
870                 pr_warn("mem=nopentium ignored! (only supported on x86_32)\n");
871                 return -EINVAL;
872 #endif
873         }
874 
875         userdef = 1;
876         mem_size = memparse(p, &p);
877 
878         /* Don't remove all memory when getting "mem={invalid}" parameter: */
879         if (mem_size == 0)
880                 return -EINVAL;
881 
882         e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
883 
884         return 0;
885 }
886 early_param("mem", parse_memopt);
887 
888 static int __init parse_memmap_one(char *p)
889 {
890         char *oldp;
891         u64 start_at, mem_size;
892 
893         if (!p)
894                 return -EINVAL;
895 
896         if (!strncmp(p, "exactmap", 8)) {
897 #ifdef CONFIG_CRASH_DUMP
898                 /*
899                  * If we are doing a crash dump, we still need to know
900                  * the real memory size before the original memory map is
901                  * reset.
902                  */
903                 saved_max_pfn = e820__end_of_ram_pfn();
904 #endif
905                 e820_table->nr_entries = 0;
906                 userdef = 1;
907                 return 0;
908         }
909 
910         oldp = p;
911         mem_size = memparse(p, &p);
912         if (p == oldp)
913                 return -EINVAL;
914 
915         userdef = 1;
916         if (*p == '@') {
917                 start_at = memparse(p+1, &p);
918                 e820__range_add(start_at, mem_size, E820_TYPE_RAM);
919         } else if (*p == '#') {
920                 start_at = memparse(p+1, &p);
921                 e820__range_add(start_at, mem_size, E820_TYPE_ACPI);
922         } else if (*p == '$') {
923                 start_at = memparse(p+1, &p);
924                 e820__range_add(start_at, mem_size, E820_TYPE_RESERVED);
925         } else if (*p == '!') {
926                 start_at = memparse(p+1, &p);
927                 e820__range_add(start_at, mem_size, E820_TYPE_PRAM);
928         } else if (*p == '%') {
929                 enum e820_type from = 0, to = 0;
930 
931                 start_at = memparse(p + 1, &p);
932                 if (*p == '-')
933                         from = simple_strtoull(p + 1, &p, 0);
934                 if (*p == '+')
935                         to = simple_strtoull(p + 1, &p, 0);
936                 if (*p != '\0')
937                         return -EINVAL;
938                 if (from && to)
939                         e820__range_update(start_at, mem_size, from, to);
940                 else if (to)
941                         e820__range_add(start_at, mem_size, to);
942                 else if (from)
943                         e820__range_remove(start_at, mem_size, from, 1);
944                 else
945                         e820__range_remove(start_at, mem_size, 0, 0);
946         } else {
947                 e820__range_remove(mem_size, ULLONG_MAX - mem_size, E820_TYPE_RAM, 1);
948         }
949 
950         return *p == '\0' ? 0 : -EINVAL;
951 }
952 
953 static int __init parse_memmap_opt(char *str)
954 {
955         while (str) {
956                 char *k = strchr(str, ',');
957 
958                 if (k)
959                         *k++ = 0;
960 
961                 parse_memmap_one(str);
962                 str = k;
963         }
964 
965         return 0;
966 }
967 early_param("memmap", parse_memmap_opt);
968 
969 /*
970  * Reserve all entries from the bootloader's extensible data nodes list,
971  * because if present we are going to use it later on to fetch e820
972  * entries from it:
973  */
974 void __init e820__reserve_setup_data(void)
975 {
976         struct setup_data *data;
977         u64 pa_data;
978 
979         pa_data = boot_params.hdr.setup_data;
980         if (!pa_data)
981                 return;
982 
983         while (pa_data) {
984                 data = early_memremap(pa_data, sizeof(*data));
985                 e820__range_update(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
986                 e820__range_update_kexec(pa_data, sizeof(*data)+data->len, E820_TYPE_RAM, E820_TYPE_RESERVED_KERN);
987                 pa_data = data->next;
988                 early_memunmap(data, sizeof(*data));
989         }
990 
991         e820__update_table(e820_table);
992         e820__update_table(e820_table_kexec);
993 
994         pr_info("extended physical RAM map:\n");
995         e820__print_table("reserve setup_data");
996 }
997 
998 /*
999  * Called after parse_early_param(), after early parameters (such as mem=)
1000  * have been processed, in which case we already have an E820 table filled in
1001  * via the parameter callback function(s), but it's not sorted and printed yet:
1002  */
1003 void __init e820__finish_early_params(void)
1004 {
1005         if (userdef) {
1006                 if (e820__update_table(e820_table) < 0)
1007                         early_panic("Invalid user supplied memory map");
1008 
1009                 pr_info("user-defined physical RAM map:\n");
1010                 e820__print_table("user");
1011         }
1012 }
1013 
1014 static const char *__init e820_type_to_string(struct e820_entry *entry)
1015 {
1016         switch (entry->type) {
1017         case E820_TYPE_RESERVED_KERN:   /* Fall-through: */
1018         case E820_TYPE_RAM:             return "System RAM";
1019         case E820_TYPE_ACPI:            return "ACPI Tables";
1020         case E820_TYPE_NVS:             return "ACPI Non-volatile Storage";
1021         case E820_TYPE_UNUSABLE:        return "Unusable memory";
1022         case E820_TYPE_PRAM:            return "Persistent Memory (legacy)";
1023         case E820_TYPE_PMEM:            return "Persistent Memory";
1024         case E820_TYPE_RESERVED:        return "Reserved";
1025         default:                        return "Unknown E820 type";
1026         }
1027 }
1028 
1029 static unsigned long __init e820_type_to_iomem_type(struct e820_entry *entry)
1030 {
1031         switch (entry->type) {
1032         case E820_TYPE_RESERVED_KERN:   /* Fall-through: */
1033         case E820_TYPE_RAM:             return IORESOURCE_SYSTEM_RAM;
1034         case E820_TYPE_ACPI:            /* Fall-through: */
1035         case E820_TYPE_NVS:             /* Fall-through: */
1036         case E820_TYPE_UNUSABLE:        /* Fall-through: */
1037         case E820_TYPE_PRAM:            /* Fall-through: */
1038         case E820_TYPE_PMEM:            /* Fall-through: */
1039         case E820_TYPE_RESERVED:        /* Fall-through: */
1040         default:                        return IORESOURCE_MEM;
1041         }
1042 }
1043 
1044 static unsigned long __init e820_type_to_iores_desc(struct e820_entry *entry)
1045 {
1046         switch (entry->type) {
1047         case E820_TYPE_ACPI:            return IORES_DESC_ACPI_TABLES;
1048         case E820_TYPE_NVS:             return IORES_DESC_ACPI_NV_STORAGE;
1049         case E820_TYPE_PMEM:            return IORES_DESC_PERSISTENT_MEMORY;
1050         case E820_TYPE_PRAM:            return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
1051         case E820_TYPE_RESERVED_KERN:   /* Fall-through: */
1052         case E820_TYPE_RAM:             /* Fall-through: */
1053         case E820_TYPE_UNUSABLE:        /* Fall-through: */
1054         case E820_TYPE_RESERVED:        /* Fall-through: */
1055         default:                        return IORES_DESC_NONE;
1056         }
1057 }
1058 
1059 static bool __init do_mark_busy(enum e820_type type, struct resource *res)
1060 {
1061         /* this is the legacy bios/dos rom-shadow + mmio region */
1062         if (res->start < (1ULL<<20))
1063                 return true;
1064 
1065         /*
1066          * Treat persistent memory like device memory, i.e. reserve it
1067          * for exclusive use of a driver
1068          */
1069         switch (type) {
1070         case E820_TYPE_RESERVED:
1071         case E820_TYPE_PRAM:
1072         case E820_TYPE_PMEM:
1073                 return false;
1074         case E820_TYPE_RESERVED_KERN:
1075         case E820_TYPE_RAM:
1076         case E820_TYPE_ACPI:
1077         case E820_TYPE_NVS:
1078         case E820_TYPE_UNUSABLE:
1079         default:
1080                 return true;
1081         }
1082 }
1083 
1084 /*
1085  * Mark E820 reserved areas as busy for the resource manager:
1086  */
1087 
1088 static struct resource __initdata *e820_res;
1089 
1090 void __init e820__reserve_resources(void)
1091 {
1092         int i;
1093         struct resource *res;
1094         u64 end;
1095 
1096         res = memblock_alloc(sizeof(*res) * e820_table->nr_entries,
1097                              SMP_CACHE_BYTES);
1098         e820_res = res;
1099 
1100         for (i = 0; i < e820_table->nr_entries; i++) {
1101                 struct e820_entry *entry = e820_table->entries + i;
1102 
1103                 end = entry->addr + entry->size - 1;
1104                 if (end != (resource_size_t)end) {
1105                         res++;
1106                         continue;
1107                 }
1108                 res->start = entry->addr;
1109                 res->end   = end;
1110                 res->name  = e820_type_to_string(entry);
1111                 res->flags = e820_type_to_iomem_type(entry);
1112                 res->desc  = e820_type_to_iores_desc(entry);
1113 
1114                 /*
1115                  * Don't register the region that could be conflicted with
1116                  * PCI device BAR resources and insert them later in
1117                  * pcibios_resource_survey():
1118                  */
1119                 if (do_mark_busy(entry->type, res)) {
1120                         res->flags |= IORESOURCE_BUSY;
1121                         insert_resource(&iomem_resource, res);
1122                 }
1123                 res++;
1124         }
1125 
1126         /* Expose the bootloader-provided memory layout to the sysfs. */
1127         for (i = 0; i < e820_table_firmware->nr_entries; i++) {
1128                 struct e820_entry *entry = e820_table_firmware->entries + i;
1129 
1130                 firmware_map_add_early(entry->addr, entry->addr + entry->size, e820_type_to_string(entry));
1131         }
1132 }
1133 
1134 /*
1135  * How much should we pad the end of RAM, depending on where it is?
1136  */
1137 static unsigned long __init ram_alignment(resource_size_t pos)
1138 {
1139         unsigned long mb = pos >> 20;
1140 
1141         /* To 64kB in the first megabyte */
1142         if (!mb)
1143                 return 64*1024;
1144 
1145         /* To 1MB in the first 16MB */
1146         if (mb < 16)
1147                 return 1024*1024;
1148 
1149         /* To 64MB for anything above that */
1150         return 64*1024*1024;
1151 }
1152 
1153 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1154 
1155 void __init e820__reserve_resources_late(void)
1156 {
1157         int i;
1158         struct resource *res;
1159 
1160         res = e820_res;
1161         for (i = 0; i < e820_table->nr_entries; i++) {
1162                 if (!res->parent && res->end)
1163                         insert_resource_expand_to_fit(&iomem_resource, res);
1164                 res++;
1165         }
1166 
1167         /*
1168          * Try to bump up RAM regions to reasonable boundaries, to
1169          * avoid stolen RAM:
1170          */
1171         for (i = 0; i < e820_table->nr_entries; i++) {
1172                 struct e820_entry *entry = &e820_table->entries[i];
1173                 u64 start, end;
1174 
1175                 if (entry->type != E820_TYPE_RAM)
1176                         continue;
1177 
1178                 start = entry->addr + entry->size;
1179                 end = round_up(start, ram_alignment(start)) - 1;
1180                 if (end > MAX_RESOURCE_SIZE)
1181                         end = MAX_RESOURCE_SIZE;
1182                 if (start >= end)
1183                         continue;
1184 
1185                 printk(KERN_DEBUG "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n", start, end);
1186                 reserve_region_with_split(&iomem_resource, start, end, "RAM buffer");
1187         }
1188 }
1189 
1190 /*
1191  * Pass the firmware (bootloader) E820 map to the kernel and process it:
1192  */
1193 char *__init e820__memory_setup_default(void)
1194 {
1195         char *who = "BIOS-e820";
1196 
1197         /*
1198          * Try to copy the BIOS-supplied E820-map.
1199          *
1200          * Otherwise fake a memory map; one section from 0k->640k,
1201          * the next section from 1mb->appropriate_mem_k
1202          */
1203         if (append_e820_table(boot_params.e820_table, boot_params.e820_entries) < 0) {
1204                 u64 mem_size;
1205 
1206                 /* Compare results from other methods and take the one that gives more RAM: */
1207                 if (boot_params.alt_mem_k < boot_params.screen_info.ext_mem_k) {
1208                         mem_size = boot_params.screen_info.ext_mem_k;
1209                         who = "BIOS-88";
1210                 } else {
1211                         mem_size = boot_params.alt_mem_k;
1212                         who = "BIOS-e801";
1213                 }
1214 
1215                 e820_table->nr_entries = 0;
1216                 e820__range_add(0, LOWMEMSIZE(), E820_TYPE_RAM);
1217                 e820__range_add(HIGH_MEMORY, mem_size << 10, E820_TYPE_RAM);
1218         }
1219 
1220         /* We just appended a lot of ranges, sanitize the table: */
1221         e820__update_table(e820_table);
1222 
1223         return who;
1224 }
1225 
1226 /*
1227  * Calls e820__memory_setup_default() in essence to pick up the firmware/bootloader
1228  * E820 map - with an optional platform quirk available for virtual platforms
1229  * to override this method of boot environment processing:
1230  */
1231 void __init e820__memory_setup(void)
1232 {
1233         char *who;
1234 
1235         /* This is a firmware interface ABI - make sure we don't break it: */
1236         BUILD_BUG_ON(sizeof(struct boot_e820_entry) != 20);
1237 
1238         who = x86_init.resources.memory_setup();
1239 
1240         memcpy(e820_table_kexec, e820_table, sizeof(*e820_table_kexec));
1241         memcpy(e820_table_firmware, e820_table, sizeof(*e820_table_firmware));
1242 
1243         pr_info("BIOS-provided physical RAM map:\n");
1244         e820__print_table(who);
1245 }
1246 
1247 void __init e820__memblock_setup(void)
1248 {
1249         int i;
1250         u64 end;
1251 
1252         /*
1253          * The bootstrap memblock region count maximum is 128 entries
1254          * (INIT_MEMBLOCK_REGIONS), but EFI might pass us more E820 entries
1255          * than that - so allow memblock resizing.
1256          *
1257          * This is safe, because this call happens pretty late during x86 setup,
1258          * so we know about reserved memory regions already. (This is important
1259          * so that memblock resizing does no stomp over reserved areas.)
1260          */
1261         memblock_allow_resize();
1262 
1263         for (i = 0; i < e820_table->nr_entries; i++) {
1264                 struct e820_entry *entry = &e820_table->entries[i];
1265 
1266                 end = entry->addr + entry->size;
1267                 if (end != (resource_size_t)end)
1268                         continue;
1269 
1270                 if (entry->type != E820_TYPE_RAM && entry->type != E820_TYPE_RESERVED_KERN)
1271                         continue;
1272 
1273                 memblock_add(entry->addr, entry->size);
1274         }
1275 
1276         /* Throw away partial pages: */
1277         memblock_trim_memory(PAGE_SIZE);
1278 
1279         memblock_dump_all();
1280 }
1281 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp