[PATCH] i386: Move e820/efi memmap walking code to e820.c
[pandora-kernel.git] / arch / i386 / kernel / e820.c
1 #include <linux/kernel.h>
2 #include <linux/types.h>
3 #include <linux/init.h>
4 #include <linux/bootmem.h>
5 #include <linux/ioport.h>
6 #include <linux/string.h>
7 #include <linux/kexec.h>
8 #include <linux/module.h>
9 #include <linux/mm.h>
10 #include <linux/efi.h>
11 #include <linux/pfn.h>
12
13 #include <asm/pgtable.h>
14 #include <asm/page.h>
15 #include <asm/e820.h>
16
17 #ifdef CONFIG_EFI
18 int efi_enabled = 0;
19 EXPORT_SYMBOL(efi_enabled);
20 #endif
21
22 struct e820map e820;
23 struct change_member {
24         struct e820entry *pbios; /* pointer to original bios entry */
25         unsigned long long addr; /* address for this change point */
26 };
27 static struct change_member change_point_list[2*E820MAX] __initdata;
28 static struct change_member *change_point[2*E820MAX] __initdata;
29 static struct e820entry *overlap_list[E820MAX] __initdata;
30 static struct e820entry new_bios[E820MAX] __initdata;
31 /* For PCI or other memory-mapped resources */
32 unsigned long pci_mem_start = 0x10000000;
33 #ifdef CONFIG_PCI
34 EXPORT_SYMBOL(pci_mem_start);
35 #endif
36 struct resource data_resource = {
37         .name   = "Kernel data",
38         .start  = 0,
39         .end    = 0,
40         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
41 };
42
43 struct resource code_resource = {
44         .name   = "Kernel code",
45         .start  = 0,
46         .end    = 0,
47         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
48 };
49
50 static struct resource system_rom_resource = {
51         .name   = "System ROM",
52         .start  = 0xf0000,
53         .end    = 0xfffff,
54         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
55 };
56
57 static struct resource extension_rom_resource = {
58         .name   = "Extension ROM",
59         .start  = 0xe0000,
60         .end    = 0xeffff,
61         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
62 };
63
64 static struct resource adapter_rom_resources[] = { {
65         .name   = "Adapter ROM",
66         .start  = 0xc8000,
67         .end    = 0,
68         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
69 }, {
70         .name   = "Adapter ROM",
71         .start  = 0,
72         .end    = 0,
73         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
74 }, {
75         .name   = "Adapter ROM",
76         .start  = 0,
77         .end    = 0,
78         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
79 }, {
80         .name   = "Adapter ROM",
81         .start  = 0,
82         .end    = 0,
83         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
84 }, {
85         .name   = "Adapter ROM",
86         .start  = 0,
87         .end    = 0,
88         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
89 }, {
90         .name   = "Adapter ROM",
91         .start  = 0,
92         .end    = 0,
93         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
94 } };
95
96 static struct resource video_rom_resource = {
97         .name   = "Video ROM",
98         .start  = 0xc0000,
99         .end    = 0xc7fff,
100         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
101 };
102
103 static struct resource video_ram_resource = {
104         .name   = "Video RAM area",
105         .start  = 0xa0000,
106         .end    = 0xbffff,
107         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
108 };
109
110 static struct resource standard_io_resources[] = { {
111         .name   = "dma1",
112         .start  = 0x0000,
113         .end    = 0x001f,
114         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
115 }, {
116         .name   = "pic1",
117         .start  = 0x0020,
118         .end    = 0x0021,
119         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
120 }, {
121         .name   = "timer0",
122         .start  = 0x0040,
123         .end    = 0x0043,
124         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
125 }, {
126         .name   = "timer1",
127         .start  = 0x0050,
128         .end    = 0x0053,
129         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
130 }, {
131         .name   = "keyboard",
132         .start  = 0x0060,
133         .end    = 0x006f,
134         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
135 }, {
136         .name   = "dma page reg",
137         .start  = 0x0080,
138         .end    = 0x008f,
139         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
140 }, {
141         .name   = "pic2",
142         .start  = 0x00a0,
143         .end    = 0x00a1,
144         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
145 }, {
146         .name   = "dma2",
147         .start  = 0x00c0,
148         .end    = 0x00df,
149         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
150 }, {
151         .name   = "fpu",
152         .start  = 0x00f0,
153         .end    = 0x00ff,
154         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
155 } };
156
157 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
158
159 static int __init romchecksum(unsigned char *rom, unsigned long length)
160 {
161         unsigned char *p, sum = 0;
162
163         for (p = rom; p < rom + length; p++)
164                 sum += *p;
165         return sum == 0;
166 }
167
168 static void __init probe_roms(void)
169 {
170         unsigned long start, length, upper;
171         unsigned char *rom;
172         int           i;
173
174         /* video rom */
175         upper = adapter_rom_resources[0].start;
176         for (start = video_rom_resource.start; start < upper; start += 2048) {
177                 rom = isa_bus_to_virt(start);
178                 if (!romsignature(rom))
179                         continue;
180
181                 video_rom_resource.start = start;
182
183                 /* 0 < length <= 0x7f * 512, historically */
184                 length = rom[2] * 512;
185
186                 /* if checksum okay, trust length byte */
187                 if (length && romchecksum(rom, length))
188                         video_rom_resource.end = start + length - 1;
189
190                 request_resource(&iomem_resource, &video_rom_resource);
191                 break;
192         }
193
194         start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
195         if (start < upper)
196                 start = upper;
197
198         /* system rom */
199         request_resource(&iomem_resource, &system_rom_resource);
200         upper = system_rom_resource.start;
201
202         /* check for extension rom (ignore length byte!) */
203         rom = isa_bus_to_virt(extension_rom_resource.start);
204         if (romsignature(rom)) {
205                 length = extension_rom_resource.end - extension_rom_resource.start + 1;
206                 if (romchecksum(rom, length)) {
207                         request_resource(&iomem_resource, &extension_rom_resource);
208                         upper = extension_rom_resource.start;
209                 }
210         }
211
212         /* check for adapter roms on 2k boundaries */
213         for (i = 0; i < ARRAY_SIZE(adapter_rom_resources) && start < upper; start += 2048) {
214                 rom = isa_bus_to_virt(start);
215                 if (!romsignature(rom))
216                         continue;
217
218                 /* 0 < length <= 0x7f * 512, historically */
219                 length = rom[2] * 512;
220
221                 /* but accept any length that fits if checksum okay */
222                 if (!length || start + length > upper || !romchecksum(rom, length))
223                         continue;
224
225                 adapter_rom_resources[i].start = start;
226                 adapter_rom_resources[i].end = start + length - 1;
227                 request_resource(&iomem_resource, &adapter_rom_resources[i]);
228
229                 start = adapter_rom_resources[i++].end & ~2047UL;
230         }
231 }
232
233 /*
234  * Request address space for all standard RAM and ROM resources
235  * and also for regions reported as reserved by the e820.
236  */
237 static void __init
238 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
239 {
240         int i;
241
242         probe_roms();
243         for (i = 0; i < e820.nr_map; i++) {
244                 struct resource *res;
245 #ifndef CONFIG_RESOURCES_64BIT
246                 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
247                         continue;
248 #endif
249                 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
250                 switch (e820.map[i].type) {
251                 case E820_RAM:  res->name = "System RAM"; break;
252                 case E820_ACPI: res->name = "ACPI Tables"; break;
253                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
254                 default:        res->name = "reserved";
255                 }
256                 res->start = e820.map[i].addr;
257                 res->end = res->start + e820.map[i].size - 1;
258                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
259                 if (request_resource(&iomem_resource, res)) {
260                         kfree(res);
261                         continue;
262                 }
263                 if (e820.map[i].type == E820_RAM) {
264                         /*
265                          *  We don't know which RAM region contains kernel data,
266                          *  so we try it repeatedly and let the resource manager
267                          *  test it.
268                          */
269                         request_resource(res, code_resource);
270                         request_resource(res, data_resource);
271 #ifdef CONFIG_KEXEC
272                         request_resource(res, &crashk_res);
273 #endif
274                 }
275         }
276 }
277
278 /*
279  * Request address space for all standard resources
280  *
281  * This is called just before pcibios_init(), which is also a
282  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
283  */
284 static int __init request_standard_resources(void)
285 {
286         int i;
287
288         printk("Setting up standard PCI resources\n");
289         if (efi_enabled)
290                 efi_initialize_iomem_resources(&code_resource, &data_resource);
291         else
292                 legacy_init_iomem_resources(&code_resource, &data_resource);
293
294         /* EFI systems may still have VGA */
295         request_resource(&iomem_resource, &video_ram_resource);
296
297         /* request I/O space for devices used on all i[345]86 PCs */
298         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
299                 request_resource(&ioport_resource, &standard_io_resources[i]);
300         return 0;
301 }
302
303 subsys_initcall(request_standard_resources);
304
305 void __init add_memory_region(unsigned long long start,
306                               unsigned long long size, int type)
307 {
308         int x;
309
310         if (!efi_enabled) {
311                 x = e820.nr_map;
312
313                 if (x == E820MAX) {
314                     printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
315                     return;
316                 }
317
318                 e820.map[x].addr = start;
319                 e820.map[x].size = size;
320                 e820.map[x].type = type;
321                 e820.nr_map++;
322         }
323 } /* add_memory_region */
324
325 /*
326  * Sanitize the BIOS e820 map.
327  *
328  * Some e820 responses include overlapping entries.  The following
329  * replaces the original e820 map with a new one, removing overlaps.
330  *
331  */
332 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
333 {
334         struct change_member *change_tmp;
335         unsigned long current_type, last_type;
336         unsigned long long last_addr;
337         int chgidx, still_changing;
338         int overlap_entries;
339         int new_bios_entry;
340         int old_nr, new_nr, chg_nr;
341         int i;
342
343         /*
344                 Visually we're performing the following (1,2,3,4 = memory types)...
345
346                 Sample memory map (w/overlaps):
347                    ____22__________________
348                    ______________________4_
349                    ____1111________________
350                    _44_____________________
351                    11111111________________
352                    ____________________33__
353                    ___________44___________
354                    __________33333_________
355                    ______________22________
356                    ___________________2222_
357                    _________111111111______
358                    _____________________11_
359                    _________________4______
360
361                 Sanitized equivalent (no overlap):
362                    1_______________________
363                    _44_____________________
364                    ___1____________________
365                    ____22__________________
366                    ______11________________
367                    _________1______________
368                    __________3_____________
369                    ___________44___________
370                    _____________33_________
371                    _______________2________
372                    ________________1_______
373                    _________________4______
374                    ___________________2____
375                    ____________________33__
376                    ______________________4_
377         */
378         printk("sanitize start\n");
379         /* if there's only one memory region, don't bother */
380         if (*pnr_map < 2) {
381                 printk("sanitize bail 0\n");
382                 return -1;
383         }
384
385         old_nr = *pnr_map;
386
387         /* bail out if we find any unreasonable addresses in bios map */
388         for (i=0; i<old_nr; i++)
389                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) {
390                         printk("sanitize bail 1\n");
391                         return -1;
392                 }
393
394         /* create pointers for initial change-point information (for sorting) */
395         for (i=0; i < 2*old_nr; i++)
396                 change_point[i] = &change_point_list[i];
397
398         /* record all known change-points (starting and ending addresses),
399            omitting those that are for empty memory regions */
400         chgidx = 0;
401         for (i=0; i < old_nr; i++)      {
402                 if (biosmap[i].size != 0) {
403                         change_point[chgidx]->addr = biosmap[i].addr;
404                         change_point[chgidx++]->pbios = &biosmap[i];
405                         change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
406                         change_point[chgidx++]->pbios = &biosmap[i];
407                 }
408         }
409         chg_nr = chgidx;        /* true number of change-points */
410
411         /* sort change-point list by memory addresses (low -> high) */
412         still_changing = 1;
413         while (still_changing)  {
414                 still_changing = 0;
415                 for (i=1; i < chg_nr; i++)  {
416                         /* if <current_addr> > <last_addr>, swap */
417                         /* or, if current=<start_addr> & last=<end_addr>, swap */
418                         if ((change_point[i]->addr < change_point[i-1]->addr) ||
419                                 ((change_point[i]->addr == change_point[i-1]->addr) &&
420                                  (change_point[i]->addr == change_point[i]->pbios->addr) &&
421                                  (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
422                            )
423                         {
424                                 change_tmp = change_point[i];
425                                 change_point[i] = change_point[i-1];
426                                 change_point[i-1] = change_tmp;
427                                 still_changing=1;
428                         }
429                 }
430         }
431
432         /* create a new bios memory map, removing overlaps */
433         overlap_entries=0;       /* number of entries in the overlap table */
434         new_bios_entry=0;        /* index for creating new bios map entries */
435         last_type = 0;           /* start with undefined memory type */
436         last_addr = 0;           /* start with 0 as last starting address */
437         /* loop through change-points, determining affect on the new bios map */
438         for (chgidx=0; chgidx < chg_nr; chgidx++)
439         {
440                 /* keep track of all overlapping bios entries */
441                 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
442                 {
443                         /* add map entry to overlap list (> 1 entry implies an overlap) */
444                         overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
445                 }
446                 else
447                 {
448                         /* remove entry from list (order independent, so swap with last) */
449                         for (i=0; i<overlap_entries; i++)
450                         {
451                                 if (overlap_list[i] == change_point[chgidx]->pbios)
452                                         overlap_list[i] = overlap_list[overlap_entries-1];
453                         }
454                         overlap_entries--;
455                 }
456                 /* if there are overlapping entries, decide which "type" to use */
457                 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
458                 current_type = 0;
459                 for (i=0; i<overlap_entries; i++)
460                         if (overlap_list[i]->type > current_type)
461                                 current_type = overlap_list[i]->type;
462                 /* continue building up new bios map based on this information */
463                 if (current_type != last_type)  {
464                         if (last_type != 0)      {
465                                 new_bios[new_bios_entry].size =
466                                         change_point[chgidx]->addr - last_addr;
467                                 /* move forward only if the new size was non-zero */
468                                 if (new_bios[new_bios_entry].size != 0)
469                                         if (++new_bios_entry >= E820MAX)
470                                                 break;  /* no more space left for new bios entries */
471                         }
472                         if (current_type != 0)  {
473                                 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
474                                 new_bios[new_bios_entry].type = current_type;
475                                 last_addr=change_point[chgidx]->addr;
476                         }
477                         last_type = current_type;
478                 }
479         }
480         new_nr = new_bios_entry;   /* retain count for new bios entries */
481
482         /* copy new bios mapping into original location */
483         memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
484         *pnr_map = new_nr;
485
486         printk("sanitize end\n");
487         return 0;
488 }
489
490 /*
491  * Copy the BIOS e820 map into a safe place.
492  *
493  * Sanity-check it while we're at it..
494  *
495  * If we're lucky and live on a modern system, the setup code
496  * will have given us a memory map that we can use to properly
497  * set up memory.  If we aren't, we'll fake a memory map.
498  *
499  * We check to see that the memory map contains at least 2 elements
500  * before we'll use it, because the detection code in setup.S may
501  * not be perfect and most every PC known to man has two memory
502  * regions: one from 0 to 640k, and one from 1mb up.  (The IBM
503  * thinkpad 560x, for example, does not cooperate with the memory
504  * detection code.)
505  */
506 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
507 {
508         /* Only one memory region (or negative)? Ignore it */
509         if (nr_map < 2)
510                 return -1;
511
512         do {
513                 unsigned long long start = biosmap->addr;
514                 unsigned long long size = biosmap->size;
515                 unsigned long long end = start + size;
516                 unsigned long type = biosmap->type;
517                 printk("copy_e820_map() start: %016Lx size: %016Lx end: %016Lx type: %ld\n", start, size, end, type);
518
519                 /* Overflow in 64 bits? Ignore the memory map. */
520                 if (start > end)
521                         return -1;
522
523                 /*
524                  * Some BIOSes claim RAM in the 640k - 1M region.
525                  * Not right. Fix it up.
526                  */
527                 if (type == E820_RAM) {
528                         printk("copy_e820_map() type is E820_RAM\n");
529                         if (start < 0x100000ULL && end > 0xA0000ULL) {
530                                 printk("copy_e820_map() lies in range...\n");
531                                 if (start < 0xA0000ULL) {
532                                         printk("copy_e820_map() start < 0xA0000ULL\n");
533                                         add_memory_region(start, 0xA0000ULL-start, type);
534                                 }
535                                 if (end <= 0x100000ULL) {
536                                         printk("copy_e820_map() end <= 0x100000ULL\n");
537                                         continue;
538                                 }
539                                 start = 0x100000ULL;
540                                 size = end - start;
541                         }
542                 }
543                 add_memory_region(start, size, type);
544         } while (biosmap++,--nr_map);
545         return 0;
546 }
547
548 /*
549  * Callback for efi_memory_walk.
550  */
551 static int __init
552 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
553 {
554         unsigned long *max_pfn = arg, pfn;
555
556         if (start < end) {
557                 pfn = PFN_UP(end -1);
558                 if (pfn > *max_pfn)
559                         *max_pfn = pfn;
560         }
561         return 0;
562 }
563
564 static int __init
565 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
566 {
567         memory_present(0, PFN_UP(start), PFN_DOWN(end));
568         return 0;
569 }
570
571 /*
572  * Find the highest page frame number we have available
573  */
574 void __init find_max_pfn(void)
575 {
576         int i;
577
578         max_pfn = 0;
579         if (efi_enabled) {
580                 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
581                 efi_memmap_walk(efi_memory_present_wrapper, NULL);
582                 return;
583         }
584
585         for (i = 0; i < e820.nr_map; i++) {
586                 unsigned long start, end;
587                 /* RAM? */
588                 if (e820.map[i].type != E820_RAM)
589                         continue;
590                 start = PFN_UP(e820.map[i].addr);
591                 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
592                 if (start >= end)
593                         continue;
594                 if (end > max_pfn)
595                         max_pfn = end;
596                 memory_present(0, start, end);
597         }
598 }
599
600 /*
601  * Free all available memory for boot time allocation.  Used
602  * as a callback function by efi_memory_walk()
603  */
604
605 static int __init
606 free_available_memory(unsigned long start, unsigned long end, void *arg)
607 {
608         /* check max_low_pfn */
609         if (start >= (max_low_pfn << PAGE_SHIFT))
610                 return 0;
611         if (end >= (max_low_pfn << PAGE_SHIFT))
612                 end = max_low_pfn << PAGE_SHIFT;
613         if (start < end)
614                 free_bootmem(start, end - start);
615
616         return 0;
617 }
618 /*
619  * Register fully available low RAM pages with the bootmem allocator.
620  */
621 void __init register_bootmem_low_pages(unsigned long max_low_pfn)
622 {
623         int i;
624
625         if (efi_enabled) {
626                 efi_memmap_walk(free_available_memory, NULL);
627                 return;
628         }
629         for (i = 0; i < e820.nr_map; i++) {
630                 unsigned long curr_pfn, last_pfn, size;
631                 /*
632                  * Reserve usable low memory
633                  */
634                 if (e820.map[i].type != E820_RAM)
635                         continue;
636                 /*
637                  * We are rounding up the start address of usable memory:
638                  */
639                 curr_pfn = PFN_UP(e820.map[i].addr);
640                 if (curr_pfn >= max_low_pfn)
641                         continue;
642                 /*
643                  * ... and at the end of the usable range downwards:
644                  */
645                 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
646
647                 if (last_pfn > max_low_pfn)
648                         last_pfn = max_low_pfn;
649
650                 /*
651                  * .. finally, did all the rounding and playing
652                  * around just make the area go away?
653                  */
654                 if (last_pfn <= curr_pfn)
655                         continue;
656
657                 size = last_pfn - curr_pfn;
658                 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
659         }
660 }
661
662 void __init register_memory(void)
663 {
664         unsigned long gapstart, gapsize, round;
665         unsigned long long last;
666         int i;
667
668         /*
669          * Search for the bigest gap in the low 32 bits of the e820
670          * memory space.
671          */
672         last = 0x100000000ull;
673         gapstart = 0x10000000;
674         gapsize = 0x400000;
675         i = e820.nr_map;
676         while (--i >= 0) {
677                 unsigned long long start = e820.map[i].addr;
678                 unsigned long long end = start + e820.map[i].size;
679
680                 /*
681                  * Since "last" is at most 4GB, we know we'll
682                  * fit in 32 bits if this condition is true
683                  */
684                 if (last > end) {
685                         unsigned long gap = last - end;
686
687                         if (gap > gapsize) {
688                                 gapsize = gap;
689                                 gapstart = end;
690                         }
691                 }
692                 if (start < last)
693                         last = start;
694         }
695
696         /*
697          * See how much we want to round up: start off with
698          * rounding to the next 1MB area.
699          */
700         round = 0x100000;
701         while ((gapsize >> 4) > round)
702                 round += round;
703         /* Fun with two's complement */
704         pci_mem_start = (gapstart + round) & -round;
705
706         printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
707                 pci_mem_start, gapstart, gapsize);
708 }