]> Pileus Git - ~andy/linux/blob - arch/x86/platform/efi/efi.c
rbd: drop an unsafe assertion
[~andy/linux] / arch / x86 / platform / efi / 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  * Copyright (C) 2013 SuSE Labs
16  *      Borislav Petkov <bp@suse.de> - runtime services VA mapping
17  *
18  * Copied from efi_32.c to eliminate the duplicated code between EFI
19  * 32/64 support code. --ying 2007-10-26
20  *
21  * All EFI Runtime Services are not implemented yet as EFI only
22  * supports physical mode addressing on SoftSDV. This is to be fixed
23  * in a future version.  --drummond 1999-07-20
24  *
25  * Implemented EFI runtime services and virtual mode calls.  --davidm
26  *
27  * Goutham Rao: <goutham.rao@intel.com>
28  *      Skip non-WB memory and ignore empty memory ranges.
29  */
30
31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
32
33 #include <linux/kernel.h>
34 #include <linux/init.h>
35 #include <linux/efi.h>
36 #include <linux/efi-bgrt.h>
37 #include <linux/export.h>
38 #include <linux/bootmem.h>
39 #include <linux/slab.h>
40 #include <linux/memblock.h>
41 #include <linux/spinlock.h>
42 #include <linux/uaccess.h>
43 #include <linux/time.h>
44 #include <linux/io.h>
45 #include <linux/reboot.h>
46 #include <linux/bcd.h>
47
48 #include <asm/setup.h>
49 #include <asm/efi.h>
50 #include <asm/time.h>
51 #include <asm/cacheflush.h>
52 #include <asm/tlbflush.h>
53 #include <asm/x86_init.h>
54 #include <asm/rtc.h>
55
56 #define EFI_DEBUG
57
58 #define EFI_MIN_RESERVE 5120
59
60 #define EFI_DUMMY_GUID \
61         EFI_GUID(0x4424ac57, 0xbe4b, 0x47dd, 0x9e, 0x97, 0xed, 0x50, 0xf0, 0x9f, 0x92, 0xa9)
62
63 static efi_char16_t efi_dummy_name[6] = { 'D', 'U', 'M', 'M', 'Y', 0 };
64
65 struct efi_memory_map memmap;
66
67 static struct efi efi_phys __initdata;
68 static efi_system_table_t efi_systab __initdata;
69
70 unsigned long x86_efi_facility;
71
72 static __initdata efi_config_table_type_t arch_tables[] = {
73 #ifdef CONFIG_X86_UV
74         {UV_SYSTEM_TABLE_GUID, "UVsystab", &efi.uv_systab},
75 #endif
76         {NULL_GUID, NULL, NULL},
77 };
78
79 u64 efi_setup;          /* efi setup_data physical address */
80
81 /*
82  * Returns 1 if 'facility' is enabled, 0 otherwise.
83  */
84 int efi_enabled(int facility)
85 {
86         return test_bit(facility, &x86_efi_facility) != 0;
87 }
88 EXPORT_SYMBOL(efi_enabled);
89
90 static bool __initdata disable_runtime = false;
91 static int __init setup_noefi(char *arg)
92 {
93         disable_runtime = true;
94         return 0;
95 }
96 early_param("noefi", setup_noefi);
97
98 int add_efi_memmap;
99 EXPORT_SYMBOL(add_efi_memmap);
100
101 static int __init setup_add_efi_memmap(char *arg)
102 {
103         add_efi_memmap = 1;
104         return 0;
105 }
106 early_param("add_efi_memmap", setup_add_efi_memmap);
107
108 static bool efi_no_storage_paranoia;
109
110 static int __init setup_storage_paranoia(char *arg)
111 {
112         efi_no_storage_paranoia = true;
113         return 0;
114 }
115 early_param("efi_no_storage_paranoia", setup_storage_paranoia);
116
117 static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
118 {
119         unsigned long flags;
120         efi_status_t status;
121
122         spin_lock_irqsave(&rtc_lock, flags);
123         status = efi_call_virt2(get_time, tm, tc);
124         spin_unlock_irqrestore(&rtc_lock, flags);
125         return status;
126 }
127
128 static efi_status_t virt_efi_set_time(efi_time_t *tm)
129 {
130         unsigned long flags;
131         efi_status_t status;
132
133         spin_lock_irqsave(&rtc_lock, flags);
134         status = efi_call_virt1(set_time, tm);
135         spin_unlock_irqrestore(&rtc_lock, flags);
136         return status;
137 }
138
139 static efi_status_t virt_efi_get_wakeup_time(efi_bool_t *enabled,
140                                              efi_bool_t *pending,
141                                              efi_time_t *tm)
142 {
143         unsigned long flags;
144         efi_status_t status;
145
146         spin_lock_irqsave(&rtc_lock, flags);
147         status = efi_call_virt3(get_wakeup_time,
148                                 enabled, pending, tm);
149         spin_unlock_irqrestore(&rtc_lock, flags);
150         return status;
151 }
152
153 static efi_status_t virt_efi_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
154 {
155         unsigned long flags;
156         efi_status_t status;
157
158         spin_lock_irqsave(&rtc_lock, flags);
159         status = efi_call_virt2(set_wakeup_time,
160                                 enabled, tm);
161         spin_unlock_irqrestore(&rtc_lock, flags);
162         return status;
163 }
164
165 static efi_status_t virt_efi_get_variable(efi_char16_t *name,
166                                           efi_guid_t *vendor,
167                                           u32 *attr,
168                                           unsigned long *data_size,
169                                           void *data)
170 {
171         return efi_call_virt5(get_variable,
172                               name, vendor, attr,
173                               data_size, data);
174 }
175
176 static efi_status_t virt_efi_get_next_variable(unsigned long *name_size,
177                                                efi_char16_t *name,
178                                                efi_guid_t *vendor)
179 {
180         return efi_call_virt3(get_next_variable,
181                               name_size, name, vendor);
182 }
183
184 static efi_status_t virt_efi_set_variable(efi_char16_t *name,
185                                           efi_guid_t *vendor,
186                                           u32 attr,
187                                           unsigned long data_size,
188                                           void *data)
189 {
190         return efi_call_virt5(set_variable,
191                               name, vendor, attr,
192                               data_size, data);
193 }
194
195 static efi_status_t virt_efi_query_variable_info(u32 attr,
196                                                  u64 *storage_space,
197                                                  u64 *remaining_space,
198                                                  u64 *max_variable_size)
199 {
200         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
201                 return EFI_UNSUPPORTED;
202
203         return efi_call_virt4(query_variable_info, attr, storage_space,
204                               remaining_space, max_variable_size);
205 }
206
207 static efi_status_t virt_efi_get_next_high_mono_count(u32 *count)
208 {
209         return efi_call_virt1(get_next_high_mono_count, count);
210 }
211
212 static void virt_efi_reset_system(int reset_type,
213                                   efi_status_t status,
214                                   unsigned long data_size,
215                                   efi_char16_t *data)
216 {
217         efi_call_virt4(reset_system, reset_type, status,
218                        data_size, data);
219 }
220
221 static efi_status_t virt_efi_update_capsule(efi_capsule_header_t **capsules,
222                                             unsigned long count,
223                                             unsigned long sg_list)
224 {
225         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
226                 return EFI_UNSUPPORTED;
227
228         return efi_call_virt3(update_capsule, capsules, count, sg_list);
229 }
230
231 static efi_status_t virt_efi_query_capsule_caps(efi_capsule_header_t **capsules,
232                                                 unsigned long count,
233                                                 u64 *max_size,
234                                                 int *reset_type)
235 {
236         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
237                 return EFI_UNSUPPORTED;
238
239         return efi_call_virt4(query_capsule_caps, capsules, count, max_size,
240                               reset_type);
241 }
242
243 static efi_status_t __init phys_efi_set_virtual_address_map(
244         unsigned long memory_map_size,
245         unsigned long descriptor_size,
246         u32 descriptor_version,
247         efi_memory_desc_t *virtual_map)
248 {
249         efi_status_t status;
250
251         efi_call_phys_prelog();
252         status = efi_call_phys4(efi_phys.set_virtual_address_map,
253                                 memory_map_size, descriptor_size,
254                                 descriptor_version, virtual_map);
255         efi_call_phys_epilog();
256         return status;
257 }
258
259 static efi_status_t __init phys_efi_get_time(efi_time_t *tm,
260                                              efi_time_cap_t *tc)
261 {
262         unsigned long flags;
263         efi_status_t status;
264
265         spin_lock_irqsave(&rtc_lock, flags);
266         efi_call_phys_prelog();
267         status = efi_call_phys2(efi_phys.get_time, virt_to_phys(tm),
268                                 virt_to_phys(tc));
269         efi_call_phys_epilog();
270         spin_unlock_irqrestore(&rtc_lock, flags);
271         return status;
272 }
273
274 int efi_set_rtc_mmss(const struct timespec *now)
275 {
276         unsigned long nowtime = now->tv_sec;
277         efi_status_t    status;
278         efi_time_t      eft;
279         efi_time_cap_t  cap;
280         struct rtc_time tm;
281
282         status = efi.get_time(&eft, &cap);
283         if (status != EFI_SUCCESS) {
284                 pr_err("Oops: efitime: can't read time!\n");
285                 return -1;
286         }
287
288         rtc_time_to_tm(nowtime, &tm);
289         if (!rtc_valid_tm(&tm)) {
290                 eft.year = tm.tm_year + 1900;
291                 eft.month = tm.tm_mon + 1;
292                 eft.day = tm.tm_mday;
293                 eft.minute = tm.tm_min;
294                 eft.second = tm.tm_sec;
295                 eft.nanosecond = 0;
296         } else {
297                 printk(KERN_ERR
298                        "%s: Invalid EFI RTC value: write of %lx to EFI RTC failed\n",
299                        __FUNCTION__, nowtime);
300                 return -1;
301         }
302
303         status = efi.set_time(&eft);
304         if (status != EFI_SUCCESS) {
305                 pr_err("Oops: efitime: can't write time!\n");
306                 return -1;
307         }
308         return 0;
309 }
310
311 void efi_get_time(struct timespec *now)
312 {
313         efi_status_t status;
314         efi_time_t eft;
315         efi_time_cap_t cap;
316
317         status = efi.get_time(&eft, &cap);
318         if (status != EFI_SUCCESS)
319                 pr_err("Oops: efitime: can't read time!\n");
320
321         now->tv_sec = mktime(eft.year, eft.month, eft.day, eft.hour,
322                              eft.minute, eft.second);
323         now->tv_nsec = 0;
324 }
325
326 /*
327  * Tell the kernel about the EFI memory map.  This might include
328  * more than the max 128 entries that can fit in the e820 legacy
329  * (zeropage) memory map.
330  */
331
332 static void __init do_add_efi_memmap(void)
333 {
334         void *p;
335
336         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
337                 efi_memory_desc_t *md = p;
338                 unsigned long long start = md->phys_addr;
339                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
340                 int e820_type;
341
342                 switch (md->type) {
343                 case EFI_LOADER_CODE:
344                 case EFI_LOADER_DATA:
345                 case EFI_BOOT_SERVICES_CODE:
346                 case EFI_BOOT_SERVICES_DATA:
347                 case EFI_CONVENTIONAL_MEMORY:
348                         if (md->attribute & EFI_MEMORY_WB)
349                                 e820_type = E820_RAM;
350                         else
351                                 e820_type = E820_RESERVED;
352                         break;
353                 case EFI_ACPI_RECLAIM_MEMORY:
354                         e820_type = E820_ACPI;
355                         break;
356                 case EFI_ACPI_MEMORY_NVS:
357                         e820_type = E820_NVS;
358                         break;
359                 case EFI_UNUSABLE_MEMORY:
360                         e820_type = E820_UNUSABLE;
361                         break;
362                 default:
363                         /*
364                          * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
365                          * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
366                          * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
367                          */
368                         e820_type = E820_RESERVED;
369                         break;
370                 }
371                 e820_add_region(start, size, e820_type);
372         }
373         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
374 }
375
376 int __init efi_memblock_x86_reserve_range(void)
377 {
378         struct efi_info *e = &boot_params.efi_info;
379         unsigned long pmap;
380
381 #ifdef CONFIG_X86_32
382         /* Can't handle data above 4GB at this time */
383         if (e->efi_memmap_hi) {
384                 pr_err("Memory map is above 4GB, disabling EFI.\n");
385                 return -EINVAL;
386         }
387         pmap =  e->efi_memmap;
388 #else
389         pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
390 #endif
391         memmap.phys_map         = (void *)pmap;
392         memmap.nr_map           = e->efi_memmap_size /
393                                   e->efi_memdesc_size;
394         memmap.desc_size        = e->efi_memdesc_size;
395         memmap.desc_version     = e->efi_memdesc_version;
396
397         memblock_reserve(pmap, memmap.nr_map * memmap.desc_size);
398
399         efi.memmap = &memmap;
400
401         return 0;
402 }
403
404 static void __init print_efi_memmap(void)
405 {
406 #ifdef EFI_DEBUG
407         efi_memory_desc_t *md;
408         void *p;
409         int i;
410
411         for (p = memmap.map, i = 0;
412              p < memmap.map_end;
413              p += memmap.desc_size, i++) {
414                 md = p;
415                 pr_info("mem%02u: type=%u, attr=0x%llx, "
416                         "range=[0x%016llx-0x%016llx) (%lluMB)\n",
417                         i, md->type, md->attribute, md->phys_addr,
418                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
419                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
420         }
421 #endif  /*  EFI_DEBUG  */
422 }
423
424 void __init efi_reserve_boot_services(void)
425 {
426         void *p;
427
428         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
429                 efi_memory_desc_t *md = p;
430                 u64 start = md->phys_addr;
431                 u64 size = md->num_pages << EFI_PAGE_SHIFT;
432
433                 if (md->type != EFI_BOOT_SERVICES_CODE &&
434                     md->type != EFI_BOOT_SERVICES_DATA)
435                         continue;
436                 /* Only reserve where possible:
437                  * - Not within any already allocated areas
438                  * - Not over any memory area (really needed, if above?)
439                  * - Not within any part of the kernel
440                  * - Not the bios reserved area
441                 */
442                 if ((start + size > __pa_symbol(_text)
443                                 && start <= __pa_symbol(_end)) ||
444                         !e820_all_mapped(start, start+size, E820_RAM) ||
445                         memblock_is_region_reserved(start, size)) {
446                         /* Could not reserve, skip it */
447                         md->num_pages = 0;
448                         memblock_dbg("Could not reserve boot range "
449                                         "[0x%010llx-0x%010llx]\n",
450                                                 start, start+size-1);
451                 } else
452                         memblock_reserve(start, size);
453         }
454 }
455
456 void __init efi_unmap_memmap(void)
457 {
458         clear_bit(EFI_MEMMAP, &x86_efi_facility);
459         if (memmap.map) {
460                 early_iounmap(memmap.map, memmap.nr_map * memmap.desc_size);
461                 memmap.map = NULL;
462         }
463 }
464
465 void __init efi_free_boot_services(void)
466 {
467         void *p;
468
469         if (!efi_is_native())
470                 return;
471
472         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
473                 efi_memory_desc_t *md = p;
474                 unsigned long long start = md->phys_addr;
475                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
476
477                 if (md->type != EFI_BOOT_SERVICES_CODE &&
478                     md->type != EFI_BOOT_SERVICES_DATA)
479                         continue;
480
481                 /* Could not reserve boot area */
482                 if (!size)
483                         continue;
484
485                 free_bootmem_late(start, size);
486         }
487
488         efi_unmap_memmap();
489 }
490
491 static int __init efi_systab_init(void *phys)
492 {
493         if (efi_enabled(EFI_64BIT)) {
494                 efi_system_table_64_t *systab64;
495                 struct efi_setup_data *data = NULL;
496                 u64 tmp = 0;
497
498                 if (efi_setup) {
499                         data = early_memremap(efi_setup, sizeof(*data));
500                         if (!data)
501                                 return -ENOMEM;
502                 }
503                 systab64 = early_ioremap((unsigned long)phys,
504                                          sizeof(*systab64));
505                 if (systab64 == NULL) {
506                         pr_err("Couldn't map the system table!\n");
507                         if (data)
508                                 early_iounmap(data, sizeof(*data));
509                         return -ENOMEM;
510                 }
511
512                 efi_systab.hdr = systab64->hdr;
513                 efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
514                                               systab64->fw_vendor;
515                 tmp |= data ? data->fw_vendor : systab64->fw_vendor;
516                 efi_systab.fw_revision = systab64->fw_revision;
517                 efi_systab.con_in_handle = systab64->con_in_handle;
518                 tmp |= systab64->con_in_handle;
519                 efi_systab.con_in = systab64->con_in;
520                 tmp |= systab64->con_in;
521                 efi_systab.con_out_handle = systab64->con_out_handle;
522                 tmp |= systab64->con_out_handle;
523                 efi_systab.con_out = systab64->con_out;
524                 tmp |= systab64->con_out;
525                 efi_systab.stderr_handle = systab64->stderr_handle;
526                 tmp |= systab64->stderr_handle;
527                 efi_systab.stderr = systab64->stderr;
528                 tmp |= systab64->stderr;
529                 efi_systab.runtime = data ?
530                                      (void *)(unsigned long)data->runtime :
531                                      (void *)(unsigned long)systab64->runtime;
532                 tmp |= data ? data->runtime : systab64->runtime;
533                 efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
534                 tmp |= systab64->boottime;
535                 efi_systab.nr_tables = systab64->nr_tables;
536                 efi_systab.tables = data ? (unsigned long)data->tables :
537                                            systab64->tables;
538                 tmp |= data ? data->tables : systab64->tables;
539
540                 early_iounmap(systab64, sizeof(*systab64));
541                 if (data)
542                         early_iounmap(data, sizeof(*data));
543 #ifdef CONFIG_X86_32
544                 if (tmp >> 32) {
545                         pr_err("EFI data located above 4GB, disabling EFI.\n");
546                         return -EINVAL;
547                 }
548 #endif
549         } else {
550                 efi_system_table_32_t *systab32;
551
552                 systab32 = early_ioremap((unsigned long)phys,
553                                          sizeof(*systab32));
554                 if (systab32 == NULL) {
555                         pr_err("Couldn't map the system table!\n");
556                         return -ENOMEM;
557                 }
558
559                 efi_systab.hdr = systab32->hdr;
560                 efi_systab.fw_vendor = systab32->fw_vendor;
561                 efi_systab.fw_revision = systab32->fw_revision;
562                 efi_systab.con_in_handle = systab32->con_in_handle;
563                 efi_systab.con_in = systab32->con_in;
564                 efi_systab.con_out_handle = systab32->con_out_handle;
565                 efi_systab.con_out = systab32->con_out;
566                 efi_systab.stderr_handle = systab32->stderr_handle;
567                 efi_systab.stderr = systab32->stderr;
568                 efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
569                 efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
570                 efi_systab.nr_tables = systab32->nr_tables;
571                 efi_systab.tables = systab32->tables;
572
573                 early_iounmap(systab32, sizeof(*systab32));
574         }
575
576         efi.systab = &efi_systab;
577
578         /*
579          * Verify the EFI Table
580          */
581         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
582                 pr_err("System table signature incorrect!\n");
583                 return -EINVAL;
584         }
585         if ((efi.systab->hdr.revision >> 16) == 0)
586                 pr_err("Warning: System table version "
587                        "%d.%02d, expected 1.00 or greater!\n",
588                        efi.systab->hdr.revision >> 16,
589                        efi.systab->hdr.revision & 0xffff);
590
591         return 0;
592 }
593
594 static int __init efi_runtime_init(void)
595 {
596         efi_runtime_services_t *runtime;
597
598         /*
599          * Check out the runtime services table. We need to map
600          * the runtime services table so that we can grab the physical
601          * address of several of the EFI runtime functions, needed to
602          * set the firmware into virtual mode.
603          */
604         runtime = early_ioremap((unsigned long)efi.systab->runtime,
605                                 sizeof(efi_runtime_services_t));
606         if (!runtime) {
607                 pr_err("Could not map the runtime service table!\n");
608                 return -ENOMEM;
609         }
610         /*
611          * We will only need *early* access to the following
612          * two EFI runtime services before set_virtual_address_map
613          * is invoked.
614          */
615         efi_phys.get_time = (efi_get_time_t *)runtime->get_time;
616         efi_phys.set_virtual_address_map =
617                 (efi_set_virtual_address_map_t *)
618                 runtime->set_virtual_address_map;
619         /*
620          * Make efi_get_time can be called before entering
621          * virtual mode.
622          */
623         efi.get_time = phys_efi_get_time;
624         early_iounmap(runtime, sizeof(efi_runtime_services_t));
625
626         return 0;
627 }
628
629 static int __init efi_memmap_init(void)
630 {
631         /* Map the EFI memory map */
632         memmap.map = early_ioremap((unsigned long)memmap.phys_map,
633                                    memmap.nr_map * memmap.desc_size);
634         if (memmap.map == NULL) {
635                 pr_err("Could not map the memory map!\n");
636                 return -ENOMEM;
637         }
638         memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
639
640         if (add_efi_memmap)
641                 do_add_efi_memmap();
642
643         return 0;
644 }
645
646 /*
647  * A number of config table entries get remapped to virtual addresses
648  * after entering EFI virtual mode. However, the kexec kernel requires
649  * their physical addresses therefore we pass them via setup_data and
650  * correct those entries to their respective physical addresses here.
651  *
652  * Currently only handles smbios which is necessary for some firmware
653  * implementation.
654  */
655 static int __init efi_reuse_config(u64 tables, int nr_tables)
656 {
657         int i, sz, ret = 0;
658         void *p, *tablep;
659         struct efi_setup_data *data;
660
661         if (!efi_setup)
662                 return 0;
663
664         if (!efi_enabled(EFI_64BIT))
665                 return 0;
666
667         data = early_memremap(efi_setup, sizeof(*data));
668         if (!data) {
669                 ret = -ENOMEM;
670                 goto out;
671         }
672
673         if (!data->smbios)
674                 goto out_memremap;
675
676         sz = sizeof(efi_config_table_64_t);
677
678         p = tablep = early_memremap(tables, nr_tables * sz);
679         if (!p) {
680                 pr_err("Could not map Configuration table!\n");
681                 ret = -ENOMEM;
682                 goto out_memremap;
683         }
684
685         for (i = 0; i < efi.systab->nr_tables; i++) {
686                 efi_guid_t guid;
687
688                 guid = ((efi_config_table_64_t *)p)->guid;
689
690                 if (!efi_guidcmp(guid, SMBIOS_TABLE_GUID))
691                         ((efi_config_table_64_t *)p)->table = data->smbios;
692                 p += sz;
693         }
694         early_iounmap(tablep, nr_tables * sz);
695
696 out_memremap:
697         early_iounmap(data, sizeof(*data));
698 out:
699         return ret;
700 }
701
702 void __init efi_init(void)
703 {
704         efi_char16_t *c16;
705         char vendor[100] = "unknown";
706         int i = 0;
707         void *tmp;
708
709 #ifdef CONFIG_X86_32
710         if (boot_params.efi_info.efi_systab_hi ||
711             boot_params.efi_info.efi_memmap_hi) {
712                 pr_info("Table located above 4GB, disabling EFI.\n");
713                 return;
714         }
715         efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
716 #else
717         efi_phys.systab = (efi_system_table_t *)
718                           (boot_params.efi_info.efi_systab |
719                           ((__u64)boot_params.efi_info.efi_systab_hi<<32));
720 #endif
721
722         if (efi_systab_init(efi_phys.systab))
723                 return;
724
725         set_bit(EFI_SYSTEM_TABLES, &x86_efi_facility);
726
727         efi.config_table = (unsigned long)efi.systab->tables;
728         efi.fw_vendor    = (unsigned long)efi.systab->fw_vendor;
729         efi.runtime      = (unsigned long)efi.systab->runtime;
730
731         /*
732          * Show what we know for posterity
733          */
734         c16 = tmp = early_ioremap(efi.systab->fw_vendor, 2);
735         if (c16) {
736                 for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
737                         vendor[i] = *c16++;
738                 vendor[i] = '\0';
739         } else
740                 pr_err("Could not map the firmware vendor!\n");
741         early_iounmap(tmp, 2);
742
743         pr_info("EFI v%u.%.02u by %s\n",
744                 efi.systab->hdr.revision >> 16,
745                 efi.systab->hdr.revision & 0xffff, vendor);
746
747         if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
748                 return;
749
750         if (efi_config_init(arch_tables))
751                 return;
752
753         set_bit(EFI_CONFIG_TABLES, &x86_efi_facility);
754
755         /*
756          * Note: We currently don't support runtime services on an EFI
757          * that doesn't match the kernel 32/64-bit mode.
758          */
759
760         if (!efi_is_native())
761                 pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
762         else {
763                 if (disable_runtime || efi_runtime_init())
764                         return;
765                 set_bit(EFI_RUNTIME_SERVICES, &x86_efi_facility);
766         }
767         if (efi_memmap_init())
768                 return;
769
770         set_bit(EFI_MEMMAP, &x86_efi_facility);
771
772         print_efi_memmap();
773 }
774
775 void __init efi_late_init(void)
776 {
777         efi_bgrt_init();
778 }
779
780 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
781 {
782         u64 addr, npages;
783
784         addr = md->virt_addr;
785         npages = md->num_pages;
786
787         memrange_efi_to_native(&addr, &npages);
788
789         if (executable)
790                 set_memory_x(addr, npages);
791         else
792                 set_memory_nx(addr, npages);
793 }
794
795 void __init runtime_code_page_mkexec(void)
796 {
797         efi_memory_desc_t *md;
798         void *p;
799
800         /* Make EFI runtime service code area executable */
801         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
802                 md = p;
803
804                 if (md->type != EFI_RUNTIME_SERVICES_CODE)
805                         continue;
806
807                 efi_set_executable(md, true);
808         }
809 }
810
811 void efi_memory_uc(u64 addr, unsigned long size)
812 {
813         unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
814         u64 npages;
815
816         npages = round_up(size, page_shift) / page_shift;
817         memrange_efi_to_native(&addr, &npages);
818         set_memory_uc(addr, npages);
819 }
820
821 void __init old_map_region(efi_memory_desc_t *md)
822 {
823         u64 start_pfn, end_pfn, end;
824         unsigned long size;
825         void *va;
826
827         start_pfn = PFN_DOWN(md->phys_addr);
828         size      = md->num_pages << PAGE_SHIFT;
829         end       = md->phys_addr + size;
830         end_pfn   = PFN_UP(end);
831
832         if (pfn_range_is_mapped(start_pfn, end_pfn)) {
833                 va = __va(md->phys_addr);
834
835                 if (!(md->attribute & EFI_MEMORY_WB))
836                         efi_memory_uc((u64)(unsigned long)va, size);
837         } else
838                 va = efi_ioremap(md->phys_addr, size,
839                                  md->type, md->attribute);
840
841         md->virt_addr = (u64) (unsigned long) va;
842         if (!va)
843                 pr_err("ioremap of 0x%llX failed!\n",
844                        (unsigned long long)md->phys_addr);
845 }
846
847 /* Merge contiguous regions of the same type and attribute */
848 static void __init efi_merge_regions(void)
849 {
850         void *p;
851         efi_memory_desc_t *md, *prev_md = NULL;
852
853         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
854                 u64 prev_size;
855                 md = p;
856
857                 if (!prev_md) {
858                         prev_md = md;
859                         continue;
860                 }
861
862                 if (prev_md->type != md->type ||
863                     prev_md->attribute != md->attribute) {
864                         prev_md = md;
865                         continue;
866                 }
867
868                 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
869
870                 if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
871                         prev_md->num_pages += md->num_pages;
872                         md->type = EFI_RESERVED_TYPE;
873                         md->attribute = 0;
874                         continue;
875                 }
876                 prev_md = md;
877         }
878 }
879
880 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
881 {
882         unsigned long size;
883         u64 end, systab;
884
885         size = md->num_pages << EFI_PAGE_SHIFT;
886         end = md->phys_addr + size;
887         systab = (u64)(unsigned long)efi_phys.systab;
888         if (md->phys_addr <= systab && systab < end) {
889                 systab += md->virt_addr - md->phys_addr;
890                 efi.systab = (efi_system_table_t *)(unsigned long)systab;
891         }
892 }
893
894 static int __init save_runtime_map(void)
895 {
896         efi_memory_desc_t *md;
897         void *tmp, *p, *q = NULL;
898         int count = 0;
899
900         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
901                 md = p;
902
903                 if (!(md->attribute & EFI_MEMORY_RUNTIME) ||
904                     (md->type == EFI_BOOT_SERVICES_CODE) ||
905                     (md->type == EFI_BOOT_SERVICES_DATA))
906                         continue;
907                 tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL);
908                 if (!tmp)
909                         goto out;
910                 q = tmp;
911
912                 memcpy(q + count * memmap.desc_size, md, memmap.desc_size);
913                 count++;
914         }
915
916         efi_runtime_map_setup(q, count, memmap.desc_size);
917
918         return 0;
919 out:
920         kfree(q);
921         return -ENOMEM;
922 }
923
924 /*
925  * Map efi regions which were passed via setup_data. The virt_addr is a fixed
926  * addr which was used in first kernel of a kexec boot.
927  */
928 static void __init efi_map_regions_fixed(void)
929 {
930         void *p;
931         efi_memory_desc_t *md;
932
933         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
934                 md = p;
935                 efi_map_region_fixed(md); /* FIXME: add error handling */
936                 get_systab_virt_addr(md);
937         }
938
939 }
940
941 /*
942  * Map efi memory ranges for runtime serivce and update new_memmap with virtual
943  * addresses.
944  */
945 static void * __init efi_map_regions(int *count)
946 {
947         efi_memory_desc_t *md;
948         void *p, *tmp, *new_memmap = NULL;
949
950         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
951                 md = p;
952                 if (!(md->attribute & EFI_MEMORY_RUNTIME)) {
953 #ifdef CONFIG_X86_64
954                         if (md->type != EFI_BOOT_SERVICES_CODE &&
955                             md->type != EFI_BOOT_SERVICES_DATA)
956 #endif
957                                 continue;
958                 }
959
960                 efi_map_region(md);
961                 get_systab_virt_addr(md);
962
963                 tmp = krealloc(new_memmap, (*count + 1) * memmap.desc_size,
964                                GFP_KERNEL);
965                 if (!tmp)
966                         goto out;
967                 new_memmap = tmp;
968                 memcpy(new_memmap + (*count * memmap.desc_size), md,
969                        memmap.desc_size);
970                 (*count)++;
971         }
972
973         return new_memmap;
974 out:
975         kfree(new_memmap);
976         return NULL;
977 }
978
979 /*
980  * This function will switch the EFI runtime services to virtual mode.
981  * Essentially, we look through the EFI memmap and map every region that
982  * has the runtime attribute bit set in its memory descriptor into the
983  * ->trampoline_pgd page table using a top-down VA allocation scheme.
984  *
985  * The old method which used to update that memory descriptor with the
986  * virtual address obtained from ioremap() is still supported when the
987  * kernel is booted with efi=old_map on its command line. Same old
988  * method enabled the runtime services to be called without having to
989  * thunk back into physical mode for every invocation.
990  *
991  * The new method does a pagetable switch in a preemption-safe manner
992  * so that we're in a different address space when calling a runtime
993  * function. For function arguments passing we do copy the PGDs of the
994  * kernel page table into ->trampoline_pgd prior to each call.
995  *
996  * Specially for kexec boot, efi runtime maps in previous kernel should
997  * be passed in via setup_data. In that case runtime ranges will be mapped
998  * to the same virtual addresses as the first kernel.
999  */
1000 void __init efi_enter_virtual_mode(void)
1001 {
1002         efi_status_t status;
1003         void *new_memmap = NULL;
1004         int err, count = 0;
1005
1006         efi.systab = NULL;
1007
1008         /*
1009          * We don't do virtual mode, since we don't do runtime services, on
1010          * non-native EFI
1011          */
1012         if (!efi_is_native()) {
1013                 efi_unmap_memmap();
1014                 return;
1015         }
1016
1017         if (efi_setup) {
1018                 efi_map_regions_fixed();
1019         } else {
1020                 efi_merge_regions();
1021                 new_memmap = efi_map_regions(&count);
1022                 if (!new_memmap) {
1023                         pr_err("Error reallocating memory, EFI runtime non-functional!\n");
1024                         return;
1025                 }
1026         }
1027
1028         err = save_runtime_map();
1029         if (err)
1030                 pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n");
1031
1032         BUG_ON(!efi.systab);
1033
1034         efi_setup_page_tables();
1035         efi_sync_low_kernel_mappings();
1036
1037         if (!efi_setup) {
1038                 status = phys_efi_set_virtual_address_map(
1039                         memmap.desc_size * count,
1040                         memmap.desc_size,
1041                         memmap.desc_version,
1042                         (efi_memory_desc_t *)__pa(new_memmap));
1043
1044                 if (status != EFI_SUCCESS) {
1045                         pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
1046                                  status);
1047                         panic("EFI call to SetVirtualAddressMap() failed!");
1048                 }
1049         }
1050
1051         /*
1052          * Now that EFI is in virtual mode, update the function
1053          * pointers in the runtime service table to the new virtual addresses.
1054          *
1055          * Call EFI services through wrapper functions.
1056          */
1057         efi.runtime_version = efi_systab.hdr.revision;
1058         efi.get_time = virt_efi_get_time;
1059         efi.set_time = virt_efi_set_time;
1060         efi.get_wakeup_time = virt_efi_get_wakeup_time;
1061         efi.set_wakeup_time = virt_efi_set_wakeup_time;
1062         efi.get_variable = virt_efi_get_variable;
1063         efi.get_next_variable = virt_efi_get_next_variable;
1064         efi.set_variable = virt_efi_set_variable;
1065         efi.get_next_high_mono_count = virt_efi_get_next_high_mono_count;
1066         efi.reset_system = virt_efi_reset_system;
1067         efi.set_virtual_address_map = NULL;
1068         efi.query_variable_info = virt_efi_query_variable_info;
1069         efi.update_capsule = virt_efi_update_capsule;
1070         efi.query_capsule_caps = virt_efi_query_capsule_caps;
1071
1072         efi_runtime_mkexec();
1073
1074         kfree(new_memmap);
1075
1076         /* clean DUMMY object */
1077         efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
1078                          EFI_VARIABLE_NON_VOLATILE |
1079                          EFI_VARIABLE_BOOTSERVICE_ACCESS |
1080                          EFI_VARIABLE_RUNTIME_ACCESS,
1081                          0, NULL);
1082 }
1083
1084 /*
1085  * Convenience functions to obtain memory types and attributes
1086  */
1087 u32 efi_mem_type(unsigned long phys_addr)
1088 {
1089         efi_memory_desc_t *md;
1090         void *p;
1091
1092         if (!efi_enabled(EFI_MEMMAP))
1093                 return 0;
1094
1095         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1096                 md = p;
1097                 if ((md->phys_addr <= phys_addr) &&
1098                     (phys_addr < (md->phys_addr +
1099                                   (md->num_pages << EFI_PAGE_SHIFT))))
1100                         return md->type;
1101         }
1102         return 0;
1103 }
1104
1105 u64 efi_mem_attributes(unsigned long phys_addr)
1106 {
1107         efi_memory_desc_t *md;
1108         void *p;
1109
1110         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1111                 md = p;
1112                 if ((md->phys_addr <= phys_addr) &&
1113                     (phys_addr < (md->phys_addr +
1114                                   (md->num_pages << EFI_PAGE_SHIFT))))
1115                         return md->attribute;
1116         }
1117         return 0;
1118 }
1119
1120 /*
1121  * Some firmware has serious problems when using more than 50% of the EFI
1122  * variable store, i.e. it triggers bugs that can brick machines. Ensure that
1123  * we never use more than this safe limit.
1124  *
1125  * Return EFI_SUCCESS if it is safe to write 'size' bytes to the variable
1126  * store.
1127  */
1128 efi_status_t efi_query_variable_store(u32 attributes, unsigned long size)
1129 {
1130         efi_status_t status;
1131         u64 storage_size, remaining_size, max_size;
1132
1133         if (!(attributes & EFI_VARIABLE_NON_VOLATILE))
1134                 return 0;
1135
1136         status = efi.query_variable_info(attributes, &storage_size,
1137                                          &remaining_size, &max_size);
1138         if (status != EFI_SUCCESS)
1139                 return status;
1140
1141         /*
1142          * Some firmware implementations refuse to boot if there's insufficient
1143          * space in the variable store. We account for that by refusing the
1144          * write if permitting it would reduce the available space to under
1145          * 5KB. This figure was provided by Samsung, so should be safe.
1146          */
1147         if ((remaining_size - size < EFI_MIN_RESERVE) &&
1148                 !efi_no_storage_paranoia) {
1149
1150                 /*
1151                  * Triggering garbage collection may require that the firmware
1152                  * generate a real EFI_OUT_OF_RESOURCES error. We can force
1153                  * that by attempting to use more space than is available.
1154                  */
1155                 unsigned long dummy_size = remaining_size + 1024;
1156                 void *dummy = kzalloc(dummy_size, GFP_ATOMIC);
1157
1158                 if (!dummy)
1159                         return EFI_OUT_OF_RESOURCES;
1160
1161                 status = efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
1162                                           EFI_VARIABLE_NON_VOLATILE |
1163                                           EFI_VARIABLE_BOOTSERVICE_ACCESS |
1164                                           EFI_VARIABLE_RUNTIME_ACCESS,
1165                                           dummy_size, dummy);
1166
1167                 if (status == EFI_SUCCESS) {
1168                         /*
1169                          * This should have failed, so if it didn't make sure
1170                          * that we delete it...
1171                          */
1172                         efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
1173                                          EFI_VARIABLE_NON_VOLATILE |
1174                                          EFI_VARIABLE_BOOTSERVICE_ACCESS |
1175                                          EFI_VARIABLE_RUNTIME_ACCESS,
1176                                          0, dummy);
1177                 }
1178
1179                 kfree(dummy);
1180
1181                 /*
1182                  * The runtime code may now have triggered a garbage collection
1183                  * run, so check the variable info again
1184                  */
1185                 status = efi.query_variable_info(attributes, &storage_size,
1186                                                  &remaining_size, &max_size);
1187
1188                 if (status != EFI_SUCCESS)
1189                         return status;
1190
1191                 /*
1192                  * There still isn't enough room, so return an error
1193                  */
1194                 if (remaining_size - size < EFI_MIN_RESERVE)
1195                         return EFI_OUT_OF_RESOURCES;
1196         }
1197
1198         return EFI_SUCCESS;
1199 }
1200 EXPORT_SYMBOL_GPL(efi_query_variable_store);
1201
1202 static int __init parse_efi_cmdline(char *str)
1203 {
1204         if (*str == '=')
1205                 str++;
1206
1207         if (!strncmp(str, "old_map", 7))
1208                 set_bit(EFI_OLD_MEMMAP, &x86_efi_facility);
1209
1210         return 0;
1211 }
1212 early_param("efi", parse_efi_cmdline);