x86/paravirt: don't restore second return reg
[linux-2.6] / 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
46 #define EFI_DEBUG       1
47 #define PFX             "EFI: "
48
49 int efi_enabled;
50 EXPORT_SYMBOL(efi_enabled);
51
52 struct efi efi;
53 EXPORT_SYMBOL(efi);
54
55 struct efi_memory_map memmap;
56
57 static struct efi efi_phys __initdata;
58 static efi_system_table_t efi_systab __initdata;
59
60 static int __init setup_noefi(char *arg)
61 {
62         efi_enabled = 0;
63         return 0;
64 }
65 early_param("noefi", setup_noefi);
66
67 int add_efi_memmap;
68 EXPORT_SYMBOL(add_efi_memmap);
69
70 static int __init setup_add_efi_memmap(char *arg)
71 {
72         add_efi_memmap = 1;
73         return 0;
74 }
75 early_param("add_efi_memmap", setup_add_efi_memmap);
76
77
78 static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
79 {
80         return efi_call_virt2(get_time, tm, tc);
81 }
82
83 static efi_status_t virt_efi_set_time(efi_time_t *tm)
84 {
85         return efi_call_virt1(set_time, tm);
86 }
87
88 static efi_status_t virt_efi_get_wakeup_time(efi_bool_t *enabled,
89                                              efi_bool_t *pending,
90                                              efi_time_t *tm)
91 {
92         return efi_call_virt3(get_wakeup_time,
93                               enabled, pending, tm);
94 }
95
96 static efi_status_t virt_efi_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
97 {
98         return efi_call_virt2(set_wakeup_time,
99                               enabled, tm);
100 }
101
102 static efi_status_t virt_efi_get_variable(efi_char16_t *name,
103                                           efi_guid_t *vendor,
104                                           u32 *attr,
105                                           unsigned long *data_size,
106                                           void *data)
107 {
108         return efi_call_virt5(get_variable,
109                               name, vendor, attr,
110                               data_size, data);
111 }
112
113 static efi_status_t virt_efi_get_next_variable(unsigned long *name_size,
114                                                efi_char16_t *name,
115                                                efi_guid_t *vendor)
116 {
117         return efi_call_virt3(get_next_variable,
118                               name_size, name, vendor);
119 }
120
121 static efi_status_t virt_efi_set_variable(efi_char16_t *name,
122                                           efi_guid_t *vendor,
123                                           unsigned long attr,
124                                           unsigned long data_size,
125                                           void *data)
126 {
127         return efi_call_virt5(set_variable,
128                               name, vendor, attr,
129                               data_size, data);
130 }
131
132 static efi_status_t virt_efi_get_next_high_mono_count(u32 *count)
133 {
134         return efi_call_virt1(get_next_high_mono_count, count);
135 }
136
137 static void virt_efi_reset_system(int reset_type,
138                                   efi_status_t status,
139                                   unsigned long data_size,
140                                   efi_char16_t *data)
141 {
142         efi_call_virt4(reset_system, reset_type, status,
143                        data_size, data);
144 }
145
146 static efi_status_t virt_efi_set_virtual_address_map(
147         unsigned long memory_map_size,
148         unsigned long descriptor_size,
149         u32 descriptor_version,
150         efi_memory_desc_t *virtual_map)
151 {
152         return efi_call_virt4(set_virtual_address_map,
153                               memory_map_size, descriptor_size,
154                               descriptor_version, virtual_map);
155 }
156
157 static efi_status_t __init phys_efi_set_virtual_address_map(
158         unsigned long memory_map_size,
159         unsigned long descriptor_size,
160         u32 descriptor_version,
161         efi_memory_desc_t *virtual_map)
162 {
163         efi_status_t status;
164
165         efi_call_phys_prelog();
166         status = efi_call_phys4(efi_phys.set_virtual_address_map,
167                                 memory_map_size, descriptor_size,
168                                 descriptor_version, virtual_map);
169         efi_call_phys_epilog();
170         return status;
171 }
172
173 static efi_status_t __init phys_efi_get_time(efi_time_t *tm,
174                                              efi_time_cap_t *tc)
175 {
176         efi_status_t status;
177
178         efi_call_phys_prelog();
179         status = efi_call_phys2(efi_phys.get_time, tm, tc);
180         efi_call_phys_epilog();
181         return status;
182 }
183
184 int efi_set_rtc_mmss(unsigned long nowtime)
185 {
186         int real_seconds, real_minutes;
187         efi_status_t    status;
188         efi_time_t      eft;
189         efi_time_cap_t  cap;
190
191         status = efi.get_time(&eft, &cap);
192         if (status != EFI_SUCCESS) {
193                 printk(KERN_ERR "Oops: efitime: can't read time!\n");
194                 return -1;
195         }
196
197         real_seconds = nowtime % 60;
198         real_minutes = nowtime / 60;
199         if (((abs(real_minutes - eft.minute) + 15)/30) & 1)
200                 real_minutes += 30;
201         real_minutes %= 60;
202         eft.minute = real_minutes;
203         eft.second = real_seconds;
204
205         status = efi.set_time(&eft);
206         if (status != EFI_SUCCESS) {
207                 printk(KERN_ERR "Oops: efitime: can't write time!\n");
208                 return -1;
209         }
210         return 0;
211 }
212
213 unsigned long efi_get_time(void)
214 {
215         efi_status_t status;
216         efi_time_t eft;
217         efi_time_cap_t cap;
218
219         status = efi.get_time(&eft, &cap);
220         if (status != EFI_SUCCESS)
221                 printk(KERN_ERR "Oops: efitime: can't read time!\n");
222
223         return mktime(eft.year, eft.month, eft.day, eft.hour,
224                       eft.minute, eft.second);
225 }
226
227 /*
228  * Tell the kernel about the EFI memory map.  This might include
229  * more than the max 128 entries that can fit in the e820 legacy
230  * (zeropage) memory map.
231  */
232
233 static void __init do_add_efi_memmap(void)
234 {
235         void *p;
236
237         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
238                 efi_memory_desc_t *md = p;
239                 unsigned long long start = md->phys_addr;
240                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
241                 int e820_type;
242
243                 if (md->attribute & EFI_MEMORY_WB)
244                         e820_type = E820_RAM;
245                 else
246                         e820_type = E820_RESERVED;
247                 e820_add_region(start, size, e820_type);
248         }
249         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
250 }
251
252 void __init efi_reserve_early(void)
253 {
254         unsigned long pmap;
255
256 #ifdef CONFIG_X86_32
257         pmap = boot_params.efi_info.efi_memmap;
258 #else
259         pmap = (boot_params.efi_info.efi_memmap |
260                 ((__u64)boot_params.efi_info.efi_memmap_hi<<32));
261 #endif
262         memmap.phys_map = (void *)pmap;
263         memmap.nr_map = boot_params.efi_info.efi_memmap_size /
264                 boot_params.efi_info.efi_memdesc_size;
265         memmap.desc_version = boot_params.efi_info.efi_memdesc_version;
266         memmap.desc_size = boot_params.efi_info.efi_memdesc_size;
267         reserve_early(pmap, pmap + memmap.nr_map * memmap.desc_size,
268                       "EFI memmap");
269 }
270
271 #if EFI_DEBUG
272 static void __init print_efi_memmap(void)
273 {
274         efi_memory_desc_t *md;
275         void *p;
276         int i;
277
278         for (p = memmap.map, i = 0;
279              p < memmap.map_end;
280              p += memmap.desc_size, i++) {
281                 md = p;
282                 printk(KERN_INFO PFX "mem%02u: type=%u, attr=0x%llx, "
283                         "range=[0x%016llx-0x%016llx) (%lluMB)\n",
284                         i, md->type, md->attribute, md->phys_addr,
285                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
286                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
287         }
288 }
289 #endif  /*  EFI_DEBUG  */
290
291 void __init efi_init(void)
292 {
293         efi_config_table_t *config_tables;
294         efi_runtime_services_t *runtime;
295         efi_char16_t *c16;
296         char vendor[100] = "unknown";
297         int i = 0;
298         void *tmp;
299
300 #ifdef CONFIG_X86_32
301         efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
302 #else
303         efi_phys.systab = (efi_system_table_t *)
304                 (boot_params.efi_info.efi_systab |
305                  ((__u64)boot_params.efi_info.efi_systab_hi<<32));
306 #endif
307
308         efi.systab = early_ioremap((unsigned long)efi_phys.systab,
309                                    sizeof(efi_system_table_t));
310         if (efi.systab == NULL)
311                 printk(KERN_ERR "Couldn't map the EFI system table!\n");
312         memcpy(&efi_systab, efi.systab, sizeof(efi_system_table_t));
313         early_iounmap(efi.systab, sizeof(efi_system_table_t));
314         efi.systab = &efi_systab;
315
316         /*
317          * Verify the EFI Table
318          */
319         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
320                 printk(KERN_ERR "EFI system table signature incorrect!\n");
321         if ((efi.systab->hdr.revision >> 16) == 0)
322                 printk(KERN_ERR "Warning: EFI system table version "
323                        "%d.%02d, expected 1.00 or greater!\n",
324                        efi.systab->hdr.revision >> 16,
325                        efi.systab->hdr.revision & 0xffff);
326
327         /*
328          * Show what we know for posterity
329          */
330         c16 = tmp = early_ioremap(efi.systab->fw_vendor, 2);
331         if (c16) {
332                 for (i = 0; i < sizeof(vendor) && *c16; ++i)
333                         vendor[i] = *c16++;
334                 vendor[i] = '\0';
335         } else
336                 printk(KERN_ERR PFX "Could not map the firmware vendor!\n");
337         early_iounmap(tmp, 2);
338
339         printk(KERN_INFO "EFI v%u.%.02u by %s \n",
340                efi.systab->hdr.revision >> 16,
341                efi.systab->hdr.revision & 0xffff, vendor);
342
343         /*
344          * Let's see what config tables the firmware passed to us.
345          */
346         config_tables = early_ioremap(
347                 efi.systab->tables,
348                 efi.systab->nr_tables * sizeof(efi_config_table_t));
349         if (config_tables == NULL)
350                 printk(KERN_ERR "Could not map EFI Configuration Table!\n");
351
352         printk(KERN_INFO);
353         for (i = 0; i < efi.systab->nr_tables; i++) {
354                 if (!efi_guidcmp(config_tables[i].guid, MPS_TABLE_GUID)) {
355                         efi.mps = config_tables[i].table;
356                         printk(" MPS=0x%lx ", config_tables[i].table);
357                 } else if (!efi_guidcmp(config_tables[i].guid,
358                                         ACPI_20_TABLE_GUID)) {
359                         efi.acpi20 = config_tables[i].table;
360                         printk(" ACPI 2.0=0x%lx ", config_tables[i].table);
361                 } else if (!efi_guidcmp(config_tables[i].guid,
362                                         ACPI_TABLE_GUID)) {
363                         efi.acpi = config_tables[i].table;
364                         printk(" ACPI=0x%lx ", config_tables[i].table);
365                 } else if (!efi_guidcmp(config_tables[i].guid,
366                                         SMBIOS_TABLE_GUID)) {
367                         efi.smbios = config_tables[i].table;
368                         printk(" SMBIOS=0x%lx ", config_tables[i].table);
369 #ifdef CONFIG_X86_UV
370                 } else if (!efi_guidcmp(config_tables[i].guid,
371                                         UV_SYSTEM_TABLE_GUID)) {
372                         efi.uv_systab = config_tables[i].table;
373                         printk(" UVsystab=0x%lx ", config_tables[i].table);
374 #endif
375                 } else if (!efi_guidcmp(config_tables[i].guid,
376                                         HCDP_TABLE_GUID)) {
377                         efi.hcdp = config_tables[i].table;
378                         printk(" HCDP=0x%lx ", config_tables[i].table);
379                 } else if (!efi_guidcmp(config_tables[i].guid,
380                                         UGA_IO_PROTOCOL_GUID)) {
381                         efi.uga = config_tables[i].table;
382                         printk(" UGA=0x%lx ", config_tables[i].table);
383                 }
384         }
385         printk("\n");
386         early_iounmap(config_tables,
387                           efi.systab->nr_tables * sizeof(efi_config_table_t));
388
389         /*
390          * Check out the runtime services table. We need to map
391          * the runtime services table so that we can grab the physical
392          * address of several of the EFI runtime functions, needed to
393          * set the firmware into virtual mode.
394          */
395         runtime = early_ioremap((unsigned long)efi.systab->runtime,
396                                 sizeof(efi_runtime_services_t));
397         if (runtime != NULL) {
398                 /*
399                  * We will only need *early* access to the following
400                  * two EFI runtime services before set_virtual_address_map
401                  * is invoked.
402                  */
403                 efi_phys.get_time = (efi_get_time_t *)runtime->get_time;
404                 efi_phys.set_virtual_address_map =
405                         (efi_set_virtual_address_map_t *)
406                         runtime->set_virtual_address_map;
407                 /*
408                  * Make efi_get_time can be called before entering
409                  * virtual mode.
410                  */
411                 efi.get_time = phys_efi_get_time;
412         } else
413                 printk(KERN_ERR "Could not map the EFI runtime service "
414                        "table!\n");
415         early_iounmap(runtime, sizeof(efi_runtime_services_t));
416
417         /* Map the EFI memory map */
418         memmap.map = early_ioremap((unsigned long)memmap.phys_map,
419                                    memmap.nr_map * memmap.desc_size);
420         if (memmap.map == NULL)
421                 printk(KERN_ERR "Could not map the EFI memory map!\n");
422         memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
423
424         if (memmap.desc_size != sizeof(efi_memory_desc_t))
425                 printk(KERN_WARNING
426                   "Kernel-defined memdesc doesn't match the one from EFI!\n");
427
428         if (add_efi_memmap)
429                 do_add_efi_memmap();
430
431         /* Setup for EFI runtime service */
432         reboot_type = BOOT_EFI;
433
434 #if EFI_DEBUG
435         print_efi_memmap();
436 #endif
437 }
438
439 static void __init runtime_code_page_mkexec(void)
440 {
441         efi_memory_desc_t *md;
442         void *p;
443         u64 addr, npages;
444
445         /* Make EFI runtime service code area executable */
446         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
447                 md = p;
448
449                 if (md->type != EFI_RUNTIME_SERVICES_CODE)
450                         continue;
451
452                 addr = md->virt_addr;
453                 npages = md->num_pages;
454                 memrange_efi_to_native(&addr, &npages);
455                 set_memory_x(addr, npages);
456         }
457 }
458
459 /*
460  * This function will switch the EFI runtime services to virtual mode.
461  * Essentially, look through the EFI memmap and map every region that
462  * has the runtime attribute bit set in its memory descriptor and update
463  * that memory descriptor with the virtual address obtained from ioremap().
464  * This enables the runtime services to be called without having to
465  * thunk back into physical mode for every invocation.
466  */
467 void __init efi_enter_virtual_mode(void)
468 {
469         efi_memory_desc_t *md;
470         efi_status_t status;
471         unsigned long size;
472         u64 end, systab, addr, npages;
473         void *p, *va;
474
475         efi.systab = NULL;
476         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
477                 md = p;
478                 if (!(md->attribute & EFI_MEMORY_RUNTIME))
479                         continue;
480
481                 size = md->num_pages << EFI_PAGE_SHIFT;
482                 end = md->phys_addr + size;
483
484                 if (PFN_UP(end) <= max_low_pfn_mapped)
485                         va = __va(md->phys_addr);
486                 else
487                         va = efi_ioremap(md->phys_addr, size);
488
489                 md->virt_addr = (u64) (unsigned long) va;
490
491                 if (!va) {
492                         printk(KERN_ERR PFX "ioremap of 0x%llX failed!\n",
493                                (unsigned long long)md->phys_addr);
494                         continue;
495                 }
496
497                 if (!(md->attribute & EFI_MEMORY_WB)) {
498                         addr = md->virt_addr;
499                         npages = md->num_pages;
500                         memrange_efi_to_native(&addr, &npages);
501                         set_memory_uc(addr, npages);
502                 }
503
504                 systab = (u64) (unsigned long) efi_phys.systab;
505                 if (md->phys_addr <= systab && systab < end) {
506                         systab += md->virt_addr - md->phys_addr;
507                         efi.systab = (efi_system_table_t *) (unsigned long) systab;
508                 }
509         }
510
511         BUG_ON(!efi.systab);
512
513         status = phys_efi_set_virtual_address_map(
514                 memmap.desc_size * memmap.nr_map,
515                 memmap.desc_size,
516                 memmap.desc_version,
517                 memmap.phys_map);
518
519         if (status != EFI_SUCCESS) {
520                 printk(KERN_ALERT "Unable to switch EFI into virtual mode "
521                        "(status=%lx)!\n", status);
522                 panic("EFI call to SetVirtualAddressMap() failed!");
523         }
524
525         /*
526          * Now that EFI is in virtual mode, update the function
527          * pointers in the runtime service table to the new virtual addresses.
528          *
529          * Call EFI services through wrapper functions.
530          */
531         efi.get_time = virt_efi_get_time;
532         efi.set_time = virt_efi_set_time;
533         efi.get_wakeup_time = virt_efi_get_wakeup_time;
534         efi.set_wakeup_time = virt_efi_set_wakeup_time;
535         efi.get_variable = virt_efi_get_variable;
536         efi.get_next_variable = virt_efi_get_next_variable;
537         efi.set_variable = virt_efi_set_variable;
538         efi.get_next_high_mono_count = virt_efi_get_next_high_mono_count;
539         efi.reset_system = virt_efi_reset_system;
540         efi.set_virtual_address_map = virt_efi_set_virtual_address_map;
541         if (__supported_pte_mask & _PAGE_NX)
542                 runtime_code_page_mkexec();
543         early_iounmap(memmap.map, memmap.nr_map * memmap.desc_size);
544         memmap.map = NULL;
545 }
546
547 /*
548  * Convenience functions to obtain memory types and attributes
549  */
550 u32 efi_mem_type(unsigned long phys_addr)
551 {
552         efi_memory_desc_t *md;
553         void *p;
554
555         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
556                 md = p;
557                 if ((md->phys_addr <= phys_addr) &&
558                     (phys_addr < (md->phys_addr +
559                                   (md->num_pages << EFI_PAGE_SHIFT))))
560                         return md->type;
561         }
562         return 0;
563 }
564
565 u64 efi_mem_attributes(unsigned long phys_addr)
566 {
567         efi_memory_desc_t *md;
568         void *p;
569
570         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
571                 md = p;
572                 if ((md->phys_addr <= phys_addr) &&
573                     (phys_addr < (md->phys_addr +
574                                   (md->num_pages << EFI_PAGE_SHIFT))))
575                         return md->attribute;
576         }
577         return 0;
578 }