kexec: add BSS to resource tree
[linux-2.6.git] / arch / x86 / kernel / e820_32.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 #include <linux/uaccess.h>
13 #include <linux/suspend.h>
14
15 #include <asm/pgtable.h>
16 #include <asm/page.h>
17 #include <asm/e820.h>
18 #include <asm/setup.h>
19
20 #ifdef CONFIG_EFI
21 int efi_enabled = 0;
22 EXPORT_SYMBOL(efi_enabled);
23 #endif
24
25 struct e820map e820;
26 struct change_member {
27         struct e820entry *pbios; /* pointer to original bios entry */
28         unsigned long long addr; /* address for this change point */
29 };
30 static struct change_member change_point_list[2*E820MAX] __initdata;
31 static struct change_member *change_point[2*E820MAX] __initdata;
32 static struct e820entry *overlap_list[E820MAX] __initdata;
33 static struct e820entry new_bios[E820MAX] __initdata;
34 /* For PCI or other memory-mapped resources */
35 unsigned long pci_mem_start = 0x10000000;
36 #ifdef CONFIG_PCI
37 EXPORT_SYMBOL(pci_mem_start);
38 #endif
39 extern int user_defined_memmap;
40 struct resource data_resource = {
41         .name   = "Kernel data",
42         .start  = 0,
43         .end    = 0,
44         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
45 };
46
47 struct resource code_resource = {
48         .name   = "Kernel code",
49         .start  = 0,
50         .end    = 0,
51         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
52 };
53
54 struct resource bss_resource = {
55         .name   = "Kernel bss",
56         .start  = 0,
57         .end    = 0,
58         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
59 };
60
61 static struct resource system_rom_resource = {
62         .name   = "System ROM",
63         .start  = 0xf0000,
64         .end    = 0xfffff,
65         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
66 };
67
68 static struct resource extension_rom_resource = {
69         .name   = "Extension ROM",
70         .start  = 0xe0000,
71         .end    = 0xeffff,
72         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
73 };
74
75 static struct resource adapter_rom_resources[] = { {
76         .name   = "Adapter ROM",
77         .start  = 0xc8000,
78         .end    = 0,
79         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
80 }, {
81         .name   = "Adapter ROM",
82         .start  = 0,
83         .end    = 0,
84         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
85 }, {
86         .name   = "Adapter ROM",
87         .start  = 0,
88         .end    = 0,
89         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
90 }, {
91         .name   = "Adapter ROM",
92         .start  = 0,
93         .end    = 0,
94         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
95 }, {
96         .name   = "Adapter ROM",
97         .start  = 0,
98         .end    = 0,
99         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
100 }, {
101         .name   = "Adapter ROM",
102         .start  = 0,
103         .end    = 0,
104         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
105 } };
106
107 static struct resource video_rom_resource = {
108         .name   = "Video ROM",
109         .start  = 0xc0000,
110         .end    = 0xc7fff,
111         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
112 };
113
114 static struct resource video_ram_resource = {
115         .name   = "Video RAM area",
116         .start  = 0xa0000,
117         .end    = 0xbffff,
118         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
119 };
120
121 static struct resource standard_io_resources[] = { {
122         .name   = "dma1",
123         .start  = 0x0000,
124         .end    = 0x001f,
125         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
126 }, {
127         .name   = "pic1",
128         .start  = 0x0020,
129         .end    = 0x0021,
130         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
131 }, {
132         .name   = "timer0",
133         .start  = 0x0040,
134         .end    = 0x0043,
135         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
136 }, {
137         .name   = "timer1",
138         .start  = 0x0050,
139         .end    = 0x0053,
140         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
141 }, {
142         .name   = "keyboard",
143         .start  = 0x0060,
144         .end    = 0x006f,
145         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
146 }, {
147         .name   = "dma page reg",
148         .start  = 0x0080,
149         .end    = 0x008f,
150         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
151 }, {
152         .name   = "pic2",
153         .start  = 0x00a0,
154         .end    = 0x00a1,
155         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
156 }, {
157         .name   = "dma2",
158         .start  = 0x00c0,
159         .end    = 0x00df,
160         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
161 }, {
162         .name   = "fpu",
163         .start  = 0x00f0,
164         .end    = 0x00ff,
165         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
166 } };
167
168 #define ROMSIGNATURE 0xaa55
169
170 static int __init romsignature(const unsigned char *rom)
171 {
172         const unsigned short * const ptr = (const unsigned short *)rom;
173         unsigned short sig;
174
175         return probe_kernel_address(ptr, sig) == 0 && sig == ROMSIGNATURE;
176 }
177
178 static int __init romchecksum(const unsigned char *rom, unsigned long length)
179 {
180         unsigned char sum, c;
181
182         for (sum = 0; length && probe_kernel_address(rom++, c) == 0; length--)
183                 sum += c;
184         return !length && !sum;
185 }
186
187 static void __init probe_roms(void)
188 {
189         const unsigned char *rom;
190         unsigned long start, length, upper;
191         unsigned char c;
192         int i;
193
194         /* video rom */
195         upper = adapter_rom_resources[0].start;
196         for (start = video_rom_resource.start; start < upper; start += 2048) {
197                 rom = isa_bus_to_virt(start);
198                 if (!romsignature(rom))
199                         continue;
200
201                 video_rom_resource.start = start;
202
203                 if (probe_kernel_address(rom + 2, c) != 0)
204                         continue;
205
206                 /* 0 < length <= 0x7f * 512, historically */
207                 length = c * 512;
208
209                 /* if checksum okay, trust length byte */
210                 if (length && romchecksum(rom, length))
211                         video_rom_resource.end = start + length - 1;
212
213                 request_resource(&iomem_resource, &video_rom_resource);
214                 break;
215         }
216
217         start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
218         if (start < upper)
219                 start = upper;
220
221         /* system rom */
222         request_resource(&iomem_resource, &system_rom_resource);
223         upper = system_rom_resource.start;
224
225         /* check for extension rom (ignore length byte!) */
226         rom = isa_bus_to_virt(extension_rom_resource.start);
227         if (romsignature(rom)) {
228                 length = extension_rom_resource.end - extension_rom_resource.start + 1;
229                 if (romchecksum(rom, length)) {
230                         request_resource(&iomem_resource, &extension_rom_resource);
231                         upper = extension_rom_resource.start;
232                 }
233         }
234
235         /* check for adapter roms on 2k boundaries */
236         for (i = 0; i < ARRAY_SIZE(adapter_rom_resources) && start < upper; start += 2048) {
237                 rom = isa_bus_to_virt(start);
238                 if (!romsignature(rom))
239                         continue;
240
241                 if (probe_kernel_address(rom + 2, c) != 0)
242                         continue;
243
244                 /* 0 < length <= 0x7f * 512, historically */
245                 length = c * 512;
246
247                 /* but accept any length that fits if checksum okay */
248                 if (!length || start + length > upper || !romchecksum(rom, length))
249                         continue;
250
251                 adapter_rom_resources[i].start = start;
252                 adapter_rom_resources[i].end = start + length - 1;
253                 request_resource(&iomem_resource, &adapter_rom_resources[i]);
254
255                 start = adapter_rom_resources[i++].end & ~2047UL;
256         }
257 }
258
259 /*
260  * Request address space for all standard RAM and ROM resources
261  * and also for regions reported as reserved by the e820.
262  */
263 static void __init
264 legacy_init_iomem_resources(struct resource *code_resource,
265                             struct resource *data_resource,
266                             struct resource *bss_resource)
267 {
268         int i;
269
270         probe_roms();
271         for (i = 0; i < e820.nr_map; i++) {
272                 struct resource *res;
273 #ifndef CONFIG_RESOURCES_64BIT
274                 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
275                         continue;
276 #endif
277                 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
278                 switch (e820.map[i].type) {
279                 case E820_RAM:  res->name = "System RAM"; break;
280                 case E820_ACPI: res->name = "ACPI Tables"; break;
281                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
282                 default:        res->name = "reserved";
283                 }
284                 res->start = e820.map[i].addr;
285                 res->end = res->start + e820.map[i].size - 1;
286                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
287                 if (request_resource(&iomem_resource, res)) {
288                         kfree(res);
289                         continue;
290                 }
291                 if (e820.map[i].type == E820_RAM) {
292                         /*
293                          *  We don't know which RAM region contains kernel data,
294                          *  so we try it repeatedly and let the resource manager
295                          *  test it.
296                          */
297                         request_resource(res, code_resource);
298                         request_resource(res, data_resource);
299                         request_resource(res, bss_resource);
300 #ifdef CONFIG_KEXEC
301                         if (crashk_res.start != crashk_res.end)
302                                 request_resource(res, &crashk_res);
303 #endif
304                 }
305         }
306 }
307
308 /*
309  * Request address space for all standard resources
310  *
311  * This is called just before pcibios_init(), which is also a
312  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
313  */
314 static int __init request_standard_resources(void)
315 {
316         int i;
317
318         printk("Setting up standard PCI resources\n");
319         if (efi_enabled)
320                 efi_initialize_iomem_resources(&code_resource,
321                                 &data_resource, &bss_resource);
322         else
323                 legacy_init_iomem_resources(&code_resource,
324                                 &data_resource, &bss_resource);
325
326         /* EFI systems may still have VGA */
327         request_resource(&iomem_resource, &video_ram_resource);
328
329         /* request I/O space for devices used on all i[345]86 PCs */
330         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
331                 request_resource(&ioport_resource, &standard_io_resources[i]);
332         return 0;
333 }
334
335 subsys_initcall(request_standard_resources);
336
337 #if defined(CONFIG_PM) && defined(CONFIG_HIBERNATION)
338 /**
339  * e820_mark_nosave_regions - Find the ranges of physical addresses that do not
340  * correspond to e820 RAM areas and mark the corresponding pages as nosave for
341  * hibernation.
342  *
343  * This function requires the e820 map to be sorted and without any
344  * overlapping entries and assumes the first e820 area to be RAM.
345  */
346 void __init e820_mark_nosave_regions(void)
347 {
348         int i;
349         unsigned long pfn;
350
351         pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
352         for (i = 1; i < e820.nr_map; i++) {
353                 struct e820entry *ei = &e820.map[i];
354
355                 if (pfn < PFN_UP(ei->addr))
356                         register_nosave_region(pfn, PFN_UP(ei->addr));
357
358                 pfn = PFN_DOWN(ei->addr + ei->size);
359                 if (ei->type != E820_RAM)
360                         register_nosave_region(PFN_UP(ei->addr), pfn);
361
362                 if (pfn >= max_low_pfn)
363                         break;
364         }
365 }
366 #endif
367
368 void __init add_memory_region(unsigned long long start,
369                               unsigned long long size, int type)
370 {
371         int x;
372
373         if (!efi_enabled) {
374                 x = e820.nr_map;
375
376                 if (x == E820MAX) {
377                     printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
378                     return;
379                 }
380
381                 e820.map[x].addr = start;
382                 e820.map[x].size = size;
383                 e820.map[x].type = type;
384                 e820.nr_map++;
385         }
386 } /* add_memory_region */
387
388 /*
389  * Sanitize the BIOS e820 map.
390  *
391  * Some e820 responses include overlapping entries.  The following
392  * replaces the original e820 map with a new one, removing overlaps.
393  *
394  */
395 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
396 {
397         struct change_member *change_tmp;
398         unsigned long current_type, last_type;
399         unsigned long long last_addr;
400         int chgidx, still_changing;
401         int overlap_entries;
402         int new_bios_entry;
403         int old_nr, new_nr, chg_nr;
404         int i;
405
406         /*
407                 Visually we're performing the following (1,2,3,4 = memory types)...
408
409                 Sample memory map (w/overlaps):
410                    ____22__________________
411                    ______________________4_
412                    ____1111________________
413                    _44_____________________
414                    11111111________________
415                    ____________________33__
416                    ___________44___________
417                    __________33333_________
418                    ______________22________
419                    ___________________2222_
420                    _________111111111______
421                    _____________________11_
422                    _________________4______
423
424                 Sanitized equivalent (no overlap):
425                    1_______________________
426                    _44_____________________
427                    ___1____________________
428                    ____22__________________
429                    ______11________________
430                    _________1______________
431                    __________3_____________
432                    ___________44___________
433                    _____________33_________
434                    _______________2________
435                    ________________1_______
436                    _________________4______
437                    ___________________2____
438                    ____________________33__
439                    ______________________4_
440         */
441         /* if there's only one memory region, don't bother */
442         if (*pnr_map < 2) {
443                 return -1;
444         }
445
446         old_nr = *pnr_map;
447
448         /* bail out if we find any unreasonable addresses in bios map */
449         for (i=0; i<old_nr; i++)
450                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) {
451                         return -1;
452                 }
453
454         /* create pointers for initial change-point information (for sorting) */
455         for (i=0; i < 2*old_nr; i++)
456                 change_point[i] = &change_point_list[i];
457
458         /* record all known change-points (starting and ending addresses),
459            omitting those that are for empty memory regions */
460         chgidx = 0;
461         for (i=0; i < old_nr; i++)      {
462                 if (biosmap[i].size != 0) {
463                         change_point[chgidx]->addr = biosmap[i].addr;
464                         change_point[chgidx++]->pbios = &biosmap[i];
465                         change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
466                         change_point[chgidx++]->pbios = &biosmap[i];
467                 }
468         }
469         chg_nr = chgidx;        /* true number of change-points */
470
471         /* sort change-point list by memory addresses (low -> high) */
472         still_changing = 1;
473         while (still_changing)  {
474                 still_changing = 0;
475                 for (i=1; i < chg_nr; i++)  {
476                         /* if <current_addr> > <last_addr>, swap */
477                         /* or, if current=<start_addr> & last=<end_addr>, swap */
478                         if ((change_point[i]->addr < change_point[i-1]->addr) ||
479                                 ((change_point[i]->addr == change_point[i-1]->addr) &&
480                                  (change_point[i]->addr == change_point[i]->pbios->addr) &&
481                                  (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
482                            )
483                         {
484                                 change_tmp = change_point[i];
485                                 change_point[i] = change_point[i-1];
486                                 change_point[i-1] = change_tmp;
487                                 still_changing=1;
488                         }
489                 }
490         }
491
492         /* create a new bios memory map, removing overlaps */
493         overlap_entries=0;       /* number of entries in the overlap table */
494         new_bios_entry=0;        /* index for creating new bios map entries */
495         last_type = 0;           /* start with undefined memory type */
496         last_addr = 0;           /* start with 0 as last starting address */
497         /* loop through change-points, determining affect on the new bios map */
498         for (chgidx=0; chgidx < chg_nr; chgidx++)
499         {
500                 /* keep track of all overlapping bios entries */
501                 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
502                 {
503                         /* add map entry to overlap list (> 1 entry implies an overlap) */
504                         overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
505                 }
506                 else
507                 {
508                         /* remove entry from list (order independent, so swap with last) */
509                         for (i=0; i<overlap_entries; i++)
510                         {
511                                 if (overlap_list[i] == change_point[chgidx]->pbios)
512                                         overlap_list[i] = overlap_list[overlap_entries-1];
513                         }
514                         overlap_entries--;
515                 }
516                 /* if there are overlapping entries, decide which "type" to use */
517                 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
518                 current_type = 0;
519                 for (i=0; i<overlap_entries; i++)
520                         if (overlap_list[i]->type > current_type)
521                                 current_type = overlap_list[i]->type;
522                 /* continue building up new bios map based on this information */
523                 if (current_type != last_type)  {
524                         if (last_type != 0)      {
525                                 new_bios[new_bios_entry].size =
526                                         change_point[chgidx]->addr - last_addr;
527                                 /* move forward only if the new size was non-zero */
528                                 if (new_bios[new_bios_entry].size != 0)
529                                         if (++new_bios_entry >= E820MAX)
530                                                 break;  /* no more space left for new bios entries */
531                         }
532                         if (current_type != 0)  {
533                                 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
534                                 new_bios[new_bios_entry].type = current_type;
535                                 last_addr=change_point[chgidx]->addr;
536                         }
537                         last_type = current_type;
538                 }
539         }
540         new_nr = new_bios_entry;   /* retain count for new bios entries */
541
542         /* copy new bios mapping into original location */
543         memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
544         *pnr_map = new_nr;
545
546         return 0;
547 }
548
549 /*
550  * Copy the BIOS e820 map into a safe place.
551  *
552  * Sanity-check it while we're at it..
553  *
554  * If we're lucky and live on a modern system, the setup code
555  * will have given us a memory map that we can use to properly
556  * set up memory.  If we aren't, we'll fake a memory map.
557  *
558  * We check to see that the memory map contains at least 2 elements
559  * before we'll use it, because the detection code in setup.S may
560  * not be perfect and most every PC known to man has two memory
561  * regions: one from 0 to 640k, and one from 1mb up.  (The IBM
562  * thinkpad 560x, for example, does not cooperate with the memory
563  * detection code.)
564  */
565 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
566 {
567         /* Only one memory region (or negative)? Ignore it */
568         if (nr_map < 2)
569                 return -1;
570
571         do {
572                 unsigned long long start = biosmap->addr;
573                 unsigned long long size = biosmap->size;
574                 unsigned long long end = start + size;
575                 unsigned long type = biosmap->type;
576
577                 /* Overflow in 64 bits? Ignore the memory map. */
578                 if (start > end)
579                         return -1;
580
581                 /*
582                  * Some BIOSes claim RAM in the 640k - 1M region.
583                  * Not right. Fix it up.
584                  */
585                 if (type == E820_RAM) {
586                         if (start < 0x100000ULL && end > 0xA0000ULL) {
587                                 if (start < 0xA0000ULL)
588                                         add_memory_region(start, 0xA0000ULL-start, type);
589                                 if (end <= 0x100000ULL)
590                                         continue;
591                                 start = 0x100000ULL;
592                                 size = end - start;
593                         }
594                 }
595                 add_memory_region(start, size, type);
596         } while (biosmap++,--nr_map);
597         return 0;
598 }
599
600 /*
601  * Callback for efi_memory_walk.
602  */
603 static int __init
604 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
605 {
606         unsigned long *max_pfn = arg, pfn;
607
608         if (start < end) {
609                 pfn = PFN_UP(end -1);
610                 if (pfn > *max_pfn)
611                         *max_pfn = pfn;
612         }
613         return 0;
614 }
615
616 static int __init
617 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
618 {
619         memory_present(0, PFN_UP(start), PFN_DOWN(end));
620         return 0;
621 }
622
623 /*
624  * Find the highest page frame number we have available
625  */
626 void __init find_max_pfn(void)
627 {
628         int i;
629
630         max_pfn = 0;
631         if (efi_enabled) {
632                 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
633                 efi_memmap_walk(efi_memory_present_wrapper, NULL);
634                 return;
635         }
636
637         for (i = 0; i < e820.nr_map; i++) {
638                 unsigned long start, end;
639                 /* RAM? */
640                 if (e820.map[i].type != E820_RAM)
641                         continue;
642                 start = PFN_UP(e820.map[i].addr);
643                 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
644                 if (start >= end)
645                         continue;
646                 if (end > max_pfn)
647                         max_pfn = end;
648                 memory_present(0, start, end);
649         }
650 }
651
652 /*
653  * Free all available memory for boot time allocation.  Used
654  * as a callback function by efi_memory_walk()
655  */
656
657 static int __init
658 free_available_memory(unsigned long start, unsigned long end, void *arg)
659 {
660         /* check max_low_pfn */
661         if (start >= (max_low_pfn << PAGE_SHIFT))
662                 return 0;
663         if (end >= (max_low_pfn << PAGE_SHIFT))
664                 end = max_low_pfn << PAGE_SHIFT;
665         if (start < end)
666                 free_bootmem(start, end - start);
667
668         return 0;
669 }
670 /*
671  * Register fully available low RAM pages with the bootmem allocator.
672  */
673 void __init register_bootmem_low_pages(unsigned long max_low_pfn)
674 {
675         int i;
676
677         if (efi_enabled) {
678                 efi_memmap_walk(free_available_memory, NULL);
679                 return;
680         }
681         for (i = 0; i < e820.nr_map; i++) {
682                 unsigned long curr_pfn, last_pfn, size;
683                 /*
684                  * Reserve usable low memory
685                  */
686                 if (e820.map[i].type != E820_RAM)
687                         continue;
688                 /*
689                  * We are rounding up the start address of usable memory:
690                  */
691                 curr_pfn = PFN_UP(e820.map[i].addr);
692                 if (curr_pfn >= max_low_pfn)
693                         continue;
694                 /*
695                  * ... and at the end of the usable range downwards:
696                  */
697                 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
698
699                 if (last_pfn > max_low_pfn)
700                         last_pfn = max_low_pfn;
701
702                 /*
703                  * .. finally, did all the rounding and playing
704                  * around just make the area go away?
705                  */
706                 if (last_pfn <= curr_pfn)
707                         continue;
708
709                 size = last_pfn - curr_pfn;
710                 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
711         }
712 }
713
714 void __init e820_register_memory(void)
715 {
716         unsigned long gapstart, gapsize, round;
717         unsigned long long last;
718         int i;
719
720         /*
721          * Search for the biggest gap in the low 32 bits of the e820
722          * memory space.
723          */
724         last = 0x100000000ull;
725         gapstart = 0x10000000;
726         gapsize = 0x400000;
727         i = e820.nr_map;
728         while (--i >= 0) {
729                 unsigned long long start = e820.map[i].addr;
730                 unsigned long long end = start + e820.map[i].size;
731
732                 /*
733                  * Since "last" is at most 4GB, we know we'll
734                  * fit in 32 bits if this condition is true
735                  */
736                 if (last > end) {
737                         unsigned long gap = last - end;
738
739                         if (gap > gapsize) {
740                                 gapsize = gap;
741                                 gapstart = end;
742                         }
743                 }
744                 if (start < last)
745                         last = start;
746         }
747
748         /*
749          * See how much we want to round up: start off with
750          * rounding to the next 1MB area.
751          */
752         round = 0x100000;
753         while ((gapsize >> 4) > round)
754                 round += round;
755         /* Fun with two's complement */
756         pci_mem_start = (gapstart + round) & -round;
757
758         printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
759                 pci_mem_start, gapstart, gapsize);
760 }
761
762 void __init print_memory_map(char *who)
763 {
764         int i;
765
766         for (i = 0; i < e820.nr_map; i++) {
767                 printk(" %s: %016Lx - %016Lx ", who,
768                         e820.map[i].addr,
769                         e820.map[i].addr + e820.map[i].size);
770                 switch (e820.map[i].type) {
771                 case E820_RAM:  printk("(usable)\n");
772                                 break;
773                 case E820_RESERVED:
774                                 printk("(reserved)\n");
775                                 break;
776                 case E820_ACPI:
777                                 printk("(ACPI data)\n");
778                                 break;
779                 case E820_NVS:
780                                 printk("(ACPI NVS)\n");
781                                 break;
782                 default:        printk("type %u\n", e820.map[i].type);
783                                 break;
784                 }
785         }
786 }
787
788 static __init __always_inline void efi_limit_regions(unsigned long long size)
789 {
790         unsigned long long current_addr = 0;
791         efi_memory_desc_t *md, *next_md;
792         void *p, *p1;
793         int i, j;
794
795         j = 0;
796         p1 = memmap.map;
797         for (p = p1, i = 0; p < memmap.map_end; p += memmap.desc_size, i++) {
798                 md = p;
799                 next_md = p1;
800                 current_addr = md->phys_addr +
801                         PFN_PHYS(md->num_pages);
802                 if (is_available_memory(md)) {
803                         if (md->phys_addr >= size) continue;
804                         memcpy(next_md, md, memmap.desc_size);
805                         if (current_addr >= size) {
806                                 next_md->num_pages -=
807                                         PFN_UP(current_addr-size);
808                         }
809                         p1 += memmap.desc_size;
810                         next_md = p1;
811                         j++;
812                 } else if ((md->attribute & EFI_MEMORY_RUNTIME) ==
813                            EFI_MEMORY_RUNTIME) {
814                         /* In order to make runtime services
815                          * available we have to include runtime
816                          * memory regions in memory map */
817                         memcpy(next_md, md, memmap.desc_size);
818                         p1 += memmap.desc_size;
819                         next_md = p1;
820                         j++;
821                 }
822         }
823         memmap.nr_map = j;
824         memmap.map_end = memmap.map +
825                 (memmap.nr_map * memmap.desc_size);
826 }
827
828 void __init limit_regions(unsigned long long size)
829 {
830         unsigned long long current_addr;
831         int i;
832
833         print_memory_map("limit_regions start");
834         if (efi_enabled) {
835                 efi_limit_regions(size);
836                 return;
837         }
838         for (i = 0; i < e820.nr_map; i++) {
839                 current_addr = e820.map[i].addr + e820.map[i].size;
840                 if (current_addr < size)
841                         continue;
842
843                 if (e820.map[i].type != E820_RAM)
844                         continue;
845
846                 if (e820.map[i].addr >= size) {
847                         /*
848                          * This region starts past the end of the
849                          * requested size, skip it completely.
850                          */
851                         e820.nr_map = i;
852                 } else {
853                         e820.nr_map = i + 1;
854                         e820.map[i].size -= current_addr - size;
855                 }
856                 print_memory_map("limit_regions endfor");
857                 return;
858         }
859         print_memory_map("limit_regions endfunc");
860 }
861
862 /*
863  * This function checks if any part of the range <start,end> is mapped
864  * with type.
865  */
866 int
867 e820_any_mapped(u64 start, u64 end, unsigned type)
868 {
869         int i;
870         for (i = 0; i < e820.nr_map; i++) {
871                 const struct e820entry *ei = &e820.map[i];
872                 if (type && ei->type != type)
873                         continue;
874                 if (ei->addr >= end || ei->addr + ei->size <= start)
875                         continue;
876                 return 1;
877         }
878         return 0;
879 }
880 EXPORT_SYMBOL_GPL(e820_any_mapped);
881
882  /*
883   * This function checks if the entire range <start,end> is mapped with type.
884   *
885   * Note: this function only works correct if the e820 table is sorted and
886   * not-overlapping, which is the case
887   */
888 int __init
889 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
890 {
891         u64 start = s;
892         u64 end = e;
893         int i;
894         for (i = 0; i < e820.nr_map; i++) {
895                 struct e820entry *ei = &e820.map[i];
896                 if (type && ei->type != type)
897                         continue;
898                 /* is the region (part) in overlap with the current region ?*/
899                 if (ei->addr >= end || ei->addr + ei->size <= start)
900                         continue;
901                 /* if the region is at the beginning of <start,end> we move
902                  * start to the end of the region since it's ok until there
903                  */
904                 if (ei->addr <= start)
905                         start = ei->addr + ei->size;
906                 /* if start is now at or beyond end, we're done, full
907                  * coverage */
908                 if (start >= end)
909                         return 1; /* we're done */
910         }
911         return 0;
912 }
913
914 static int __init parse_memmap(char *arg)
915 {
916         if (!arg)
917                 return -EINVAL;
918
919         if (strcmp(arg, "exactmap") == 0) {
920 #ifdef CONFIG_CRASH_DUMP
921                 /* If we are doing a crash dump, we
922                  * still need to know the real mem
923                  * size before original memory map is
924                  * reset.
925                  */
926                 find_max_pfn();
927                 saved_max_pfn = max_pfn;
928 #endif
929                 e820.nr_map = 0;
930                 user_defined_memmap = 1;
931         } else {
932                 /* If the user specifies memory size, we
933                  * limit the BIOS-provided memory map to
934                  * that size. exactmap can be used to specify
935                  * the exact map. mem=number can be used to
936                  * trim the existing memory map.
937                  */
938                 unsigned long long start_at, mem_size;
939
940                 mem_size = memparse(arg, &arg);
941                 if (*arg == '@') {
942                         start_at = memparse(arg+1, &arg);
943                         add_memory_region(start_at, mem_size, E820_RAM);
944                 } else if (*arg == '#') {
945                         start_at = memparse(arg+1, &arg);
946                         add_memory_region(start_at, mem_size, E820_ACPI);
947                 } else if (*arg == '$') {
948                         start_at = memparse(arg+1, &arg);
949                         add_memory_region(start_at, mem_size, E820_RESERVED);
950                 } else {
951                         limit_regions(mem_size);
952                         user_defined_memmap = 1;
953                 }
954         }
955         return 0;
956 }
957 early_param("memmap", parse_memmap);