Revert "cpusets: randomize node rotor used in cpuset_mem_spread_node()"
[pandora-kernel.git] / arch / x86 / kernel / efi.c
1 /*
2  * Common EFI (Extensible Firmware Interface) support functions
3  * Based on Extensible Firmware Interface Specification version 1.0
4  *
5  * Copyright (C) 1999 VA Linux Systems
6  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
7  * Copyright (C) 1999-2002 Hewlett-Packard Co.
8  *      David Mosberger-Tang <davidm@hpl.hp.com>
9  *      Stephane Eranian <eranian@hpl.hp.com>
10  * Copyright (C) 2005-2008 Intel Co.
11  *      Fenghua Yu <fenghua.yu@intel.com>
12  *      Bibo Mao <bibo.mao@intel.com>
13  *      Chandramouli Narayanan <mouli@linux.intel.com>
14  *      Huang Ying <ying.huang@intel.com>
15  *
16  * Copied from efi_32.c to eliminate the duplicated code between EFI
17  * 32/64 support code. --ying 2007-10-26
18  *
19  * All EFI Runtime Services are not implemented yet as EFI only
20  * supports physical mode addressing on SoftSDV. This is to be fixed
21  * in a future version.  --drummond 1999-07-20
22  *
23  * Implemented EFI runtime services and virtual mode calls.  --davidm
24  *
25  * Goutham Rao: <goutham.rao@intel.com>
26  *      Skip non-WB memory and ignore empty memory ranges.
27  */
28
29 #include <linux/kernel.h>
30 #include <linux/init.h>
31 #include <linux/efi.h>
32 #include <linux/bootmem.h>
33 #include <linux/spinlock.h>
34 #include <linux/uaccess.h>
35 #include <linux/time.h>
36 #include <linux/io.h>
37 #include <linux/reboot.h>
38 #include <linux/bcd.h>
39
40 #include <asm/setup.h>
41 #include <asm/efi.h>
42 #include <asm/time.h>
43 #include <asm/cacheflush.h>
44 #include <asm/tlbflush.h>
45 #include <asm/x86_init.h>
46
47 #define EFI_DEBUG       1
48 #define PFX             "EFI: "
49
50 int efi_enabled;
51 EXPORT_SYMBOL(efi_enabled);
52
53 struct efi efi;
54 EXPORT_SYMBOL(efi);
55
56 struct efi_memory_map memmap;
57
58 static struct efi efi_phys __initdata;
59 static efi_system_table_t efi_systab __initdata;
60
61 static int __init setup_noefi(char *arg)
62 {
63         efi_enabled = 0;
64         return 0;
65 }
66 early_param("noefi", setup_noefi);
67
68 int add_efi_memmap;
69 EXPORT_SYMBOL(add_efi_memmap);
70
71 static int __init setup_add_efi_memmap(char *arg)
72 {
73         add_efi_memmap = 1;
74         return 0;
75 }
76 early_param("add_efi_memmap", setup_add_efi_memmap);
77
78
79 static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
80 {
81         return efi_call_virt2(get_time, tm, tc);
82 }
83
84 static efi_status_t virt_efi_set_time(efi_time_t *tm)
85 {
86         return efi_call_virt1(set_time, tm);
87 }
88
89 static efi_status_t virt_efi_get_wakeup_time(efi_bool_t *enabled,
90                                              efi_bool_t *pending,
91                                              efi_time_t *tm)
92 {
93         return efi_call_virt3(get_wakeup_time,
94                               enabled, pending, tm);
95 }
96
97 static efi_status_t virt_efi_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
98 {
99         return efi_call_virt2(set_wakeup_time,
100                               enabled, tm);
101 }
102
103 static efi_status_t virt_efi_get_variable(efi_char16_t *name,
104                                           efi_guid_t *vendor,
105                                           u32 *attr,
106                                           unsigned long *data_size,
107                                           void *data)
108 {
109         return efi_call_virt5(get_variable,
110                               name, vendor, attr,
111                               data_size, data);
112 }
113
114 static efi_status_t virt_efi_get_next_variable(unsigned long *name_size,
115                                                efi_char16_t *name,
116                                                efi_guid_t *vendor)
117 {
118         return efi_call_virt3(get_next_variable,
119                               name_size, name, vendor);
120 }
121
122 static efi_status_t virt_efi_set_variable(efi_char16_t *name,
123                                           efi_guid_t *vendor,
124                                           unsigned long attr,
125                                           unsigned long data_size,
126                                           void *data)
127 {
128         return efi_call_virt5(set_variable,
129                               name, vendor, attr,
130                               data_size, data);
131 }
132
133 static efi_status_t virt_efi_get_next_high_mono_count(u32 *count)
134 {
135         return efi_call_virt1(get_next_high_mono_count, count);
136 }
137
138 static void virt_efi_reset_system(int reset_type,
139                                   efi_status_t status,
140                                   unsigned long data_size,
141                                   efi_char16_t *data)
142 {
143         efi_call_virt4(reset_system, reset_type, status,
144                        data_size, data);
145 }
146
147 static efi_status_t virt_efi_set_virtual_address_map(
148         unsigned long memory_map_size,
149         unsigned long descriptor_size,
150         u32 descriptor_version,
151         efi_memory_desc_t *virtual_map)
152 {
153         return efi_call_virt4(set_virtual_address_map,
154                               memory_map_size, descriptor_size,
155                               descriptor_version, virtual_map);
156 }
157
158 static efi_status_t __init phys_efi_set_virtual_address_map(
159         unsigned long memory_map_size,
160         unsigned long descriptor_size,
161         u32 descriptor_version,
162         efi_memory_desc_t *virtual_map)
163 {
164         efi_status_t status;
165
166         efi_call_phys_prelog();
167         status = efi_call_phys4(efi_phys.set_virtual_address_map,
168                                 memory_map_size, descriptor_size,
169                                 descriptor_version, virtual_map);
170         efi_call_phys_epilog();
171         return status;
172 }
173
174 static efi_status_t __init phys_efi_get_time(efi_time_t *tm,
175                                              efi_time_cap_t *tc)
176 {
177         efi_status_t status;
178
179         efi_call_phys_prelog();
180         status = efi_call_phys2(efi_phys.get_time, tm, tc);
181         efi_call_phys_epilog();
182         return status;
183 }
184
185 int efi_set_rtc_mmss(unsigned long nowtime)
186 {
187         int real_seconds, real_minutes;
188         efi_status_t    status;
189         efi_time_t      eft;
190         efi_time_cap_t  cap;
191
192         status = efi.get_time(&eft, &cap);
193         if (status != EFI_SUCCESS) {
194                 printk(KERN_ERR "Oops: efitime: can't read time!\n");
195                 return -1;
196         }
197
198         real_seconds = nowtime % 60;
199         real_minutes = nowtime / 60;
200         if (((abs(real_minutes - eft.minute) + 15)/30) & 1)
201                 real_minutes += 30;
202         real_minutes %= 60;
203         eft.minute = real_minutes;
204         eft.second = real_seconds;
205
206         status = efi.set_time(&eft);
207         if (status != EFI_SUCCESS) {
208                 printk(KERN_ERR "Oops: efitime: can't write time!\n");
209                 return -1;
210         }
211         return 0;
212 }
213
214 unsigned long efi_get_time(void)
215 {
216         efi_status_t status;
217         efi_time_t eft;
218         efi_time_cap_t cap;
219
220         status = efi.get_time(&eft, &cap);
221         if (status != EFI_SUCCESS)
222                 printk(KERN_ERR "Oops: efitime: can't read time!\n");
223
224         return mktime(eft.year, eft.month, eft.day, eft.hour,
225                       eft.minute, eft.second);
226 }
227
228 /*
229  * Tell the kernel about the EFI memory map.  This might include
230  * more than the max 128 entries that can fit in the e820 legacy
231  * (zeropage) memory map.
232  */
233
234 static void __init do_add_efi_memmap(void)
235 {
236         void *p;
237
238         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
239                 efi_memory_desc_t *md = p;
240                 unsigned long long start = md->phys_addr;
241                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
242                 int e820_type;
243
244                 switch (md->type) {
245                 case EFI_LOADER_CODE:
246                 case EFI_LOADER_DATA:
247                 case EFI_BOOT_SERVICES_CODE:
248                 case EFI_BOOT_SERVICES_DATA:
249                 case EFI_CONVENTIONAL_MEMORY:
250                         if (md->attribute & EFI_MEMORY_WB)
251                                 e820_type = E820_RAM;
252                         else
253                                 e820_type = E820_RESERVED;
254                         break;
255                 case EFI_ACPI_RECLAIM_MEMORY:
256                         e820_type = E820_ACPI;
257                         break;
258                 case EFI_ACPI_MEMORY_NVS:
259                         e820_type = E820_NVS;
260                         break;
261                 case EFI_UNUSABLE_MEMORY:
262                         e820_type = E820_UNUSABLE;
263                         break;
264                 default:
265                         /*
266                          * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
267                          * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
268                          * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
269                          */
270                         e820_type = E820_RESERVED;
271                         break;
272                 }
273                 e820_add_region(start, size, e820_type);
274         }
275         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
276 }
277
278 void __init efi_reserve_early(void)
279 {
280         unsigned long pmap;
281
282 #ifdef CONFIG_X86_32
283         pmap = boot_params.efi_info.efi_memmap;
284 #else
285         pmap = (boot_params.efi_info.efi_memmap |
286                 ((__u64)boot_params.efi_info.efi_memmap_hi<<32));
287 #endif
288         memmap.phys_map = (void *)pmap;
289         memmap.nr_map = boot_params.efi_info.efi_memmap_size /
290                 boot_params.efi_info.efi_memdesc_size;
291         memmap.desc_version = boot_params.efi_info.efi_memdesc_version;
292         memmap.desc_size = boot_params.efi_info.efi_memdesc_size;
293         reserve_early(pmap, pmap + memmap.nr_map * memmap.desc_size,
294                       "EFI memmap");
295 }
296
297 #if EFI_DEBUG
298 static void __init print_efi_memmap(void)
299 {
300         efi_memory_desc_t *md;
301         void *p;
302         int i;
303
304         for (p = memmap.map, i = 0;
305              p < memmap.map_end;
306              p += memmap.desc_size, i++) {
307                 md = p;
308                 printk(KERN_INFO PFX "mem%02u: type=%u, attr=0x%llx, "
309                         "range=[0x%016llx-0x%016llx) (%lluMB)\n",
310                         i, md->type, md->attribute, md->phys_addr,
311                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
312                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
313         }
314 }
315 #endif  /*  EFI_DEBUG  */
316
317 void __init efi_init(void)
318 {
319         efi_config_table_t *config_tables;
320         efi_runtime_services_t *runtime;
321         efi_char16_t *c16;
322         char vendor[100] = "unknown";
323         int i = 0;
324         void *tmp;
325
326 #ifdef CONFIG_X86_32
327         efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
328 #else
329         efi_phys.systab = (efi_system_table_t *)
330                 (boot_params.efi_info.efi_systab |
331                  ((__u64)boot_params.efi_info.efi_systab_hi<<32));
332 #endif
333
334         efi.systab = early_ioremap((unsigned long)efi_phys.systab,
335                                    sizeof(efi_system_table_t));
336         if (efi.systab == NULL)
337                 printk(KERN_ERR "Couldn't map the EFI system table!\n");
338         memcpy(&efi_systab, efi.systab, sizeof(efi_system_table_t));
339         early_iounmap(efi.systab, sizeof(efi_system_table_t));
340         efi.systab = &efi_systab;
341
342         /*
343          * Verify the EFI Table
344          */
345         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
346                 printk(KERN_ERR "EFI system table signature incorrect!\n");
347         if ((efi.systab->hdr.revision >> 16) == 0)
348                 printk(KERN_ERR "Warning: EFI system table version "
349                        "%d.%02d, expected 1.00 or greater!\n",
350                        efi.systab->hdr.revision >> 16,
351                        efi.systab->hdr.revision & 0xffff);
352
353         /*
354          * Show what we know for posterity
355          */
356         c16 = tmp = early_ioremap(efi.systab->fw_vendor, 2);
357         if (c16) {
358                 for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
359                         vendor[i] = *c16++;
360                 vendor[i] = '\0';
361         } else
362                 printk(KERN_ERR PFX "Could not map the firmware vendor!\n");
363         early_iounmap(tmp, 2);
364
365         printk(KERN_INFO "EFI v%u.%.02u by %s\n",
366                efi.systab->hdr.revision >> 16,
367                efi.systab->hdr.revision & 0xffff, vendor);
368
369         /*
370          * Let's see what config tables the firmware passed to us.
371          */
372         config_tables = early_ioremap(
373                 efi.systab->tables,
374                 efi.systab->nr_tables * sizeof(efi_config_table_t));
375         if (config_tables == NULL)
376                 printk(KERN_ERR "Could not map EFI Configuration Table!\n");
377
378         printk(KERN_INFO);
379         for (i = 0; i < efi.systab->nr_tables; i++) {
380                 if (!efi_guidcmp(config_tables[i].guid, MPS_TABLE_GUID)) {
381                         efi.mps = config_tables[i].table;
382                         printk(" MPS=0x%lx ", config_tables[i].table);
383                 } else if (!efi_guidcmp(config_tables[i].guid,
384                                         ACPI_20_TABLE_GUID)) {
385                         efi.acpi20 = config_tables[i].table;
386                         printk(" ACPI 2.0=0x%lx ", config_tables[i].table);
387                 } else if (!efi_guidcmp(config_tables[i].guid,
388                                         ACPI_TABLE_GUID)) {
389                         efi.acpi = config_tables[i].table;
390                         printk(" ACPI=0x%lx ", config_tables[i].table);
391                 } else if (!efi_guidcmp(config_tables[i].guid,
392                                         SMBIOS_TABLE_GUID)) {
393                         efi.smbios = config_tables[i].table;
394                         printk(" SMBIOS=0x%lx ", config_tables[i].table);
395 #ifdef CONFIG_X86_UV
396                 } else if (!efi_guidcmp(config_tables[i].guid,
397                                         UV_SYSTEM_TABLE_GUID)) {
398                         efi.uv_systab = config_tables[i].table;
399                         printk(" UVsystab=0x%lx ", config_tables[i].table);
400 #endif
401                 } else if (!efi_guidcmp(config_tables[i].guid,
402                                         HCDP_TABLE_GUID)) {
403                         efi.hcdp = config_tables[i].table;
404                         printk(" HCDP=0x%lx ", config_tables[i].table);
405                 } else if (!efi_guidcmp(config_tables[i].guid,
406                                         UGA_IO_PROTOCOL_GUID)) {
407                         efi.uga = config_tables[i].table;
408                         printk(" UGA=0x%lx ", config_tables[i].table);
409                 }
410         }
411         printk("\n");
412         early_iounmap(config_tables,
413                           efi.systab->nr_tables * sizeof(efi_config_table_t));
414
415         /*
416          * Check out the runtime services table. We need to map
417          * the runtime services table so that we can grab the physical
418          * address of several of the EFI runtime functions, needed to
419          * set the firmware into virtual mode.
420          */
421         runtime = early_ioremap((unsigned long)efi.systab->runtime,
422                                 sizeof(efi_runtime_services_t));
423         if (runtime != NULL) {
424                 /*
425                  * We will only need *early* access to the following
426                  * two EFI runtime services before set_virtual_address_map
427                  * is invoked.
428                  */
429                 efi_phys.get_time = (efi_get_time_t *)runtime->get_time;
430                 efi_phys.set_virtual_address_map =
431                         (efi_set_virtual_address_map_t *)
432                         runtime->set_virtual_address_map;
433                 /*
434                  * Make efi_get_time can be called before entering
435                  * virtual mode.
436                  */
437                 efi.get_time = phys_efi_get_time;
438         } else
439                 printk(KERN_ERR "Could not map the EFI runtime service "
440                        "table!\n");
441         early_iounmap(runtime, sizeof(efi_runtime_services_t));
442
443         /* Map the EFI memory map */
444         memmap.map = early_ioremap((unsigned long)memmap.phys_map,
445                                    memmap.nr_map * memmap.desc_size);
446         if (memmap.map == NULL)
447                 printk(KERN_ERR "Could not map the EFI memory map!\n");
448         memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
449
450         if (memmap.desc_size != sizeof(efi_memory_desc_t))
451                 printk(KERN_WARNING
452                   "Kernel-defined memdesc doesn't match the one from EFI!\n");
453
454         if (add_efi_memmap)
455                 do_add_efi_memmap();
456
457 #ifdef CONFIG_X86_32
458         x86_platform.get_wallclock = efi_get_time;
459         x86_platform.set_wallclock = efi_set_rtc_mmss;
460 #endif
461
462         /* Setup for EFI runtime service */
463         reboot_type = BOOT_EFI;
464
465 #if EFI_DEBUG
466         print_efi_memmap();
467 #endif
468 }
469
470 static void __init runtime_code_page_mkexec(void)
471 {
472         efi_memory_desc_t *md;
473         void *p;
474         u64 addr, npages;
475
476         /* Make EFI runtime service code area executable */
477         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
478                 md = p;
479
480                 if (md->type != EFI_RUNTIME_SERVICES_CODE)
481                         continue;
482
483                 addr = md->virt_addr;
484                 npages = md->num_pages;
485                 memrange_efi_to_native(&addr, &npages);
486                 set_memory_x(addr, npages);
487         }
488 }
489
490 /*
491  * This function will switch the EFI runtime services to virtual mode.
492  * Essentially, look through the EFI memmap and map every region that
493  * has the runtime attribute bit set in its memory descriptor and update
494  * that memory descriptor with the virtual address obtained from ioremap().
495  * This enables the runtime services to be called without having to
496  * thunk back into physical mode for every invocation.
497  */
498 void __init efi_enter_virtual_mode(void)
499 {
500         efi_memory_desc_t *md;
501         efi_status_t status;
502         unsigned long size;
503         u64 end, systab, addr, npages, end_pfn;
504         void *p, *va;
505
506         efi.systab = NULL;
507         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
508                 md = p;
509                 if (!(md->attribute & EFI_MEMORY_RUNTIME))
510                         continue;
511
512                 size = md->num_pages << EFI_PAGE_SHIFT;
513                 end = md->phys_addr + size;
514
515                 end_pfn = PFN_UP(end);
516                 if (end_pfn <= max_low_pfn_mapped
517                     || (end_pfn > (1UL << (32 - PAGE_SHIFT))
518                         && end_pfn <= max_pfn_mapped))
519                         va = __va(md->phys_addr);
520                 else
521                         va = efi_ioremap(md->phys_addr, size, md->type);
522
523                 md->virt_addr = (u64) (unsigned long) va;
524
525                 if (!va) {
526                         printk(KERN_ERR PFX "ioremap of 0x%llX failed!\n",
527                                (unsigned long long)md->phys_addr);
528                         continue;
529                 }
530
531                 if (!(md->attribute & EFI_MEMORY_WB)) {
532                         addr = md->virt_addr;
533                         npages = md->num_pages;
534                         memrange_efi_to_native(&addr, &npages);
535                         set_memory_uc(addr, npages);
536                 }
537
538                 systab = (u64) (unsigned long) efi_phys.systab;
539                 if (md->phys_addr <= systab && systab < end) {
540                         systab += md->virt_addr - md->phys_addr;
541                         efi.systab = (efi_system_table_t *) (unsigned long) systab;
542                 }
543         }
544
545         BUG_ON(!efi.systab);
546
547         status = phys_efi_set_virtual_address_map(
548                 memmap.desc_size * memmap.nr_map,
549                 memmap.desc_size,
550                 memmap.desc_version,
551                 memmap.phys_map);
552
553         if (status != EFI_SUCCESS) {
554                 printk(KERN_ALERT "Unable to switch EFI into virtual mode "
555                        "(status=%lx)!\n", status);
556                 panic("EFI call to SetVirtualAddressMap() failed!");
557         }
558
559         /*
560          * Now that EFI is in virtual mode, update the function
561          * pointers in the runtime service table to the new virtual addresses.
562          *
563          * Call EFI services through wrapper functions.
564          */
565         efi.get_time = virt_efi_get_time;
566         efi.set_time = virt_efi_set_time;
567         efi.get_wakeup_time = virt_efi_get_wakeup_time;
568         efi.set_wakeup_time = virt_efi_set_wakeup_time;
569         efi.get_variable = virt_efi_get_variable;
570         efi.get_next_variable = virt_efi_get_next_variable;
571         efi.set_variable = virt_efi_set_variable;
572         efi.get_next_high_mono_count = virt_efi_get_next_high_mono_count;
573         efi.reset_system = virt_efi_reset_system;
574         efi.set_virtual_address_map = virt_efi_set_virtual_address_map;
575         if (__supported_pte_mask & _PAGE_NX)
576                 runtime_code_page_mkexec();
577         early_iounmap(memmap.map, memmap.nr_map * memmap.desc_size);
578         memmap.map = NULL;
579 }
580
581 /*
582  * Convenience functions to obtain memory types and attributes
583  */
584 u32 efi_mem_type(unsigned long phys_addr)
585 {
586         efi_memory_desc_t *md;
587         void *p;
588
589         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
590                 md = p;
591                 if ((md->phys_addr <= phys_addr) &&
592                     (phys_addr < (md->phys_addr +
593                                   (md->num_pages << EFI_PAGE_SHIFT))))
594                         return md->type;
595         }
596         return 0;
597 }
598
599 u64 efi_mem_attributes(unsigned long phys_addr)
600 {
601         efi_memory_desc_t *md;
602         void *p;
603
604         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
605                 md = p;
606                 if ((md->phys_addr <= phys_addr) &&
607                     (phys_addr < (md->phys_addr +
608                                   (md->num_pages << EFI_PAGE_SHIFT))))
609                         return md->attribute;
610         }
611         return 0;
612 }