Merge /spare/repo/linux-2.6/
[linux-2.6] / arch / i386 / kernel / efi.c
1 /*
2  * Extensible Firmware Interface
3  *
4  * Based on Extensible Firmware Interface Specification version 1.0
5  *
6  * Copyright (C) 1999 VA Linux Systems
7  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
8  * Copyright (C) 1999-2002 Hewlett-Packard Co.
9  *      David Mosberger-Tang <davidm@hpl.hp.com>
10  *      Stephane Eranian <eranian@hpl.hp.com>
11  *
12  * All EFI Runtime Services are not implemented yet as EFI only
13  * supports physical mode addressing on SoftSDV. This is to be fixed
14  * in a future version.  --drummond 1999-07-20
15  *
16  * Implemented EFI runtime services and virtual mode calls.  --davidm
17  *
18  * Goutham Rao: <goutham.rao@intel.com>
19  *      Skip non-WB memory and ignore empty memory ranges.
20  */
21
22 #include <linux/config.h>
23 #include <linux/kernel.h>
24 #include <linux/init.h>
25 #include <linux/mm.h>
26 #include <linux/types.h>
27 #include <linux/time.h>
28 #include <linux/spinlock.h>
29 #include <linux/bootmem.h>
30 #include <linux/ioport.h>
31 #include <linux/module.h>
32 #include <linux/efi.h>
33 #include <linux/kexec.h>
34
35 #include <asm/setup.h>
36 #include <asm/io.h>
37 #include <asm/page.h>
38 #include <asm/pgtable.h>
39 #include <asm/processor.h>
40 #include <asm/desc.h>
41 #include <asm/tlbflush.h>
42
43 #define EFI_DEBUG       0
44 #define PFX             "EFI: "
45
46 extern efi_status_t asmlinkage efi_call_phys(void *, ...);
47
48 struct efi efi;
49 EXPORT_SYMBOL(efi);
50 static struct efi efi_phys;
51 struct efi_memory_map memmap;
52
53 /*
54  * We require an early boot_ioremap mapping mechanism initially
55  */
56 extern void * boot_ioremap(unsigned long, unsigned long);
57
58 /*
59  * To make EFI call EFI runtime service in physical addressing mode we need
60  * prelog/epilog before/after the invocation to disable interrupt, to
61  * claim EFI runtime service handler exclusively and to duplicate a memory in
62  * low memory space say 0 - 3G.
63  */
64
65 static unsigned long efi_rt_eflags;
66 static DEFINE_SPINLOCK(efi_rt_lock);
67 static pgd_t efi_bak_pg_dir_pointer[2];
68
69 static void efi_call_phys_prelog(void)
70 {
71         unsigned long cr4;
72         unsigned long temp;
73
74         spin_lock(&efi_rt_lock);
75         local_irq_save(efi_rt_eflags);
76
77         /*
78          * If I don't have PSE, I should just duplicate two entries in page
79          * directory. If I have PSE, I just need to duplicate one entry in
80          * page directory.
81          */
82         __asm__ __volatile__("movl %%cr4, %0":"=r"(cr4));
83
84         if (cr4 & X86_CR4_PSE) {
85                 efi_bak_pg_dir_pointer[0].pgd =
86                     swapper_pg_dir[pgd_index(0)].pgd;
87                 swapper_pg_dir[0].pgd =
88                     swapper_pg_dir[pgd_index(PAGE_OFFSET)].pgd;
89         } else {
90                 efi_bak_pg_dir_pointer[0].pgd =
91                     swapper_pg_dir[pgd_index(0)].pgd;
92                 efi_bak_pg_dir_pointer[1].pgd =
93                     swapper_pg_dir[pgd_index(0x400000)].pgd;
94                 swapper_pg_dir[pgd_index(0)].pgd =
95                     swapper_pg_dir[pgd_index(PAGE_OFFSET)].pgd;
96                 temp = PAGE_OFFSET + 0x400000;
97                 swapper_pg_dir[pgd_index(0x400000)].pgd =
98                     swapper_pg_dir[pgd_index(temp)].pgd;
99         }
100
101         /*
102          * After the lock is released, the original page table is restored.
103          */
104         local_flush_tlb();
105
106         cpu_gdt_descr[0].address = __pa(cpu_gdt_descr[0].address);
107         __asm__ __volatile__("lgdt %0":"=m"
108                             (*(struct Xgt_desc_struct *) __pa(&cpu_gdt_descr[0])));
109 }
110
111 static void efi_call_phys_epilog(void)
112 {
113         unsigned long cr4;
114
115         cpu_gdt_descr[0].address =
116                 (unsigned long) __va(cpu_gdt_descr[0].address);
117         __asm__ __volatile__("lgdt %0":"=m"(cpu_gdt_descr));
118         __asm__ __volatile__("movl %%cr4, %0":"=r"(cr4));
119
120         if (cr4 & X86_CR4_PSE) {
121                 swapper_pg_dir[pgd_index(0)].pgd =
122                     efi_bak_pg_dir_pointer[0].pgd;
123         } else {
124                 swapper_pg_dir[pgd_index(0)].pgd =
125                     efi_bak_pg_dir_pointer[0].pgd;
126                 swapper_pg_dir[pgd_index(0x400000)].pgd =
127                     efi_bak_pg_dir_pointer[1].pgd;
128         }
129
130         /*
131          * After the lock is released, the original page table is restored.
132          */
133         local_flush_tlb();
134
135         local_irq_restore(efi_rt_eflags);
136         spin_unlock(&efi_rt_lock);
137 }
138
139 static efi_status_t
140 phys_efi_set_virtual_address_map(unsigned long memory_map_size,
141                                  unsigned long descriptor_size,
142                                  u32 descriptor_version,
143                                  efi_memory_desc_t *virtual_map)
144 {
145         efi_status_t status;
146
147         efi_call_phys_prelog();
148         status = efi_call_phys(efi_phys.set_virtual_address_map,
149                                      memory_map_size, descriptor_size,
150                                      descriptor_version, virtual_map);
151         efi_call_phys_epilog();
152         return status;
153 }
154
155 static efi_status_t
156 phys_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
157 {
158         efi_status_t status;
159
160         efi_call_phys_prelog();
161         status = efi_call_phys(efi_phys.get_time, tm, tc);
162         efi_call_phys_epilog();
163         return status;
164 }
165
166 inline int efi_set_rtc_mmss(unsigned long nowtime)
167 {
168         int real_seconds, real_minutes;
169         efi_status_t    status;
170         efi_time_t      eft;
171         efi_time_cap_t  cap;
172
173         spin_lock(&efi_rt_lock);
174         status = efi.get_time(&eft, &cap);
175         spin_unlock(&efi_rt_lock);
176         if (status != EFI_SUCCESS)
177                 panic("Ooops, efitime: can't read time!\n");
178         real_seconds = nowtime % 60;
179         real_minutes = nowtime / 60;
180
181         if (((abs(real_minutes - eft.minute) + 15)/30) & 1)
182                 real_minutes += 30;
183         real_minutes %= 60;
184
185         eft.minute = real_minutes;
186         eft.second = real_seconds;
187
188         if (status != EFI_SUCCESS) {
189                 printk("Ooops: efitime: can't read time!\n");
190                 return -1;
191         }
192         return 0;
193 }
194 /*
195  * This should only be used during kernel init and before runtime
196  * services have been remapped, therefore, we'll need to call in physical
197  * mode.  Note, this call isn't used later, so mark it __init.
198  */
199 inline unsigned long __init efi_get_time(void)
200 {
201         efi_status_t status;
202         efi_time_t eft;
203         efi_time_cap_t cap;
204
205         status = phys_efi_get_time(&eft, &cap);
206         if (status != EFI_SUCCESS)
207                 printk("Oops: efitime: can't read time status: 0x%lx\n",status);
208
209         return mktime(eft.year, eft.month, eft.day, eft.hour,
210                         eft.minute, eft.second);
211 }
212
213 int is_available_memory(efi_memory_desc_t * md)
214 {
215         if (!(md->attribute & EFI_MEMORY_WB))
216                 return 0;
217
218         switch (md->type) {
219                 case EFI_LOADER_CODE:
220                 case EFI_LOADER_DATA:
221                 case EFI_BOOT_SERVICES_CODE:
222                 case EFI_BOOT_SERVICES_DATA:
223                 case EFI_CONVENTIONAL_MEMORY:
224                         return 1;
225         }
226         return 0;
227 }
228
229 /*
230  * We need to map the EFI memory map again after paging_init().
231  */
232 void __init efi_map_memmap(void)
233 {
234         memmap.map = NULL;
235
236         memmap.map = (efi_memory_desc_t *)
237                 bt_ioremap((unsigned long) memmap.phys_map,
238                         (memmap.nr_map * sizeof(efi_memory_desc_t)));
239
240         if (memmap.map == NULL)
241                 printk(KERN_ERR PFX "Could not remap the EFI memmap!\n");
242 }
243
244 #if EFI_DEBUG
245 static void __init print_efi_memmap(void)
246 {
247         efi_memory_desc_t *md;
248         int i;
249
250         for (i = 0; i < memmap.nr_map; i++) {
251                 md = &memmap.map[i];
252                 printk(KERN_INFO "mem%02u: type=%u, attr=0x%llx, "
253                         "range=[0x%016llx-0x%016llx) (%lluMB)\n",
254                         i, md->type, md->attribute, md->phys_addr,
255                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
256                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
257         }
258 }
259 #endif  /*  EFI_DEBUG  */
260
261 /*
262  * Walks the EFI memory map and calls CALLBACK once for each EFI
263  * memory descriptor that has memory that is available for kernel use.
264  */
265 void efi_memmap_walk(efi_freemem_callback_t callback, void *arg)
266 {
267         int prev_valid = 0;
268         struct range {
269                 unsigned long start;
270                 unsigned long end;
271         } prev, curr;
272         efi_memory_desc_t *md;
273         unsigned long start, end;
274         int i;
275
276         for (i = 0; i < memmap.nr_map; i++) {
277                 md = &memmap.map[i];
278
279                 if ((md->num_pages == 0) || (!is_available_memory(md)))
280                         continue;
281
282                 curr.start = md->phys_addr;
283                 curr.end = curr.start + (md->num_pages << EFI_PAGE_SHIFT);
284
285                 if (!prev_valid) {
286                         prev = curr;
287                         prev_valid = 1;
288                 } else {
289                         if (curr.start < prev.start)
290                                 printk(KERN_INFO PFX "Unordered memory map\n");
291                         if (prev.end == curr.start)
292                                 prev.end = curr.end;
293                         else {
294                                 start =
295                                     (unsigned long) (PAGE_ALIGN(prev.start));
296                                 end = (unsigned long) (prev.end & PAGE_MASK);
297                                 if ((end > start)
298                                     && (*callback) (start, end, arg) < 0)
299                                         return;
300                                 prev = curr;
301                         }
302                 }
303         }
304         if (prev_valid) {
305                 start = (unsigned long) PAGE_ALIGN(prev.start);
306                 end = (unsigned long) (prev.end & PAGE_MASK);
307                 if (end > start)
308                         (*callback) (start, end, arg);
309         }
310 }
311
312 void __init efi_init(void)
313 {
314         efi_config_table_t *config_tables;
315         efi_runtime_services_t *runtime;
316         efi_char16_t *c16;
317         char vendor[100] = "unknown";
318         unsigned long num_config_tables;
319         int i = 0;
320
321         memset(&efi, 0, sizeof(efi) );
322         memset(&efi_phys, 0, sizeof(efi_phys));
323
324         efi_phys.systab = EFI_SYSTAB;
325         memmap.phys_map = EFI_MEMMAP;
326         memmap.nr_map = EFI_MEMMAP_SIZE/EFI_MEMDESC_SIZE;
327         memmap.desc_version = EFI_MEMDESC_VERSION;
328
329         efi.systab = (efi_system_table_t *)
330                 boot_ioremap((unsigned long) efi_phys.systab,
331                         sizeof(efi_system_table_t));
332         /*
333          * Verify the EFI Table
334          */
335         if (efi.systab == NULL)
336                 printk(KERN_ERR PFX "Woah! Couldn't map the EFI system table.\n");
337         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
338                 printk(KERN_ERR PFX "Woah! EFI system table signature incorrect\n");
339         if ((efi.systab->hdr.revision ^ EFI_SYSTEM_TABLE_REVISION) >> 16 != 0)
340                 printk(KERN_ERR PFX
341                        "Warning: EFI system table major version mismatch: "
342                        "got %d.%02d, expected %d.%02d\n",
343                        efi.systab->hdr.revision >> 16,
344                        efi.systab->hdr.revision & 0xffff,
345                        EFI_SYSTEM_TABLE_REVISION >> 16,
346                        EFI_SYSTEM_TABLE_REVISION & 0xffff);
347         /*
348          * Grab some details from the system table
349          */
350         num_config_tables = efi.systab->nr_tables;
351         config_tables = (efi_config_table_t *)efi.systab->tables;
352         runtime = efi.systab->runtime;
353
354         /*
355          * Show what we know for posterity
356          */
357         c16 = (efi_char16_t *) boot_ioremap(efi.systab->fw_vendor, 2);
358         if (c16) {
359                 for (i = 0; i < sizeof(vendor) && *c16; ++i)
360                         vendor[i] = *c16++;
361                 vendor[i] = '\0';
362         } else
363                 printk(KERN_ERR PFX "Could not map the firmware vendor!\n");
364
365         printk(KERN_INFO PFX "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 = (efi_config_table_t *)
373                                 boot_ioremap((unsigned long) config_tables,
374                                 num_config_tables * sizeof(efi_config_table_t));
375
376         if (config_tables == NULL)
377                 printk(KERN_ERR PFX "Could not map EFI Configuration Table!\n");
378
379         for (i = 0; i < num_config_tables; i++) {
380                 if (efi_guidcmp(config_tables[i].guid, MPS_TABLE_GUID) == 0) {
381                         efi.mps = (void *)config_tables[i].table;
382                         printk(KERN_INFO " MPS=0x%lx ", config_tables[i].table);
383                 } else
384                     if (efi_guidcmp(config_tables[i].guid, ACPI_20_TABLE_GUID) == 0) {
385                         efi.acpi20 = __va(config_tables[i].table);
386                         printk(KERN_INFO " ACPI 2.0=0x%lx ", config_tables[i].table);
387                 } else
388                     if (efi_guidcmp(config_tables[i].guid, ACPI_TABLE_GUID) == 0) {
389                         efi.acpi = __va(config_tables[i].table);
390                         printk(KERN_INFO " ACPI=0x%lx ", config_tables[i].table);
391                 } else
392                     if (efi_guidcmp(config_tables[i].guid, SMBIOS_TABLE_GUID) == 0) {
393                         efi.smbios = (void *) config_tables[i].table;
394                         printk(KERN_INFO " SMBIOS=0x%lx ", config_tables[i].table);
395                 } else
396                     if (efi_guidcmp(config_tables[i].guid, HCDP_TABLE_GUID) == 0) {
397                         efi.hcdp = (void *)config_tables[i].table;
398                         printk(KERN_INFO " HCDP=0x%lx ", config_tables[i].table);
399                 } else
400                     if (efi_guidcmp(config_tables[i].guid, UGA_IO_PROTOCOL_GUID) == 0) {
401                         efi.uga = (void *)config_tables[i].table;
402                         printk(KERN_INFO " UGA=0x%lx ", config_tables[i].table);
403                 }
404         }
405         printk("\n");
406
407         /*
408          * Check out the runtime services table. We need to map
409          * the runtime services table so that we can grab the physical
410          * address of several of the EFI runtime functions, needed to
411          * set the firmware into virtual mode.
412          */
413
414         runtime = (efi_runtime_services_t *) boot_ioremap((unsigned long)
415                                                 runtime,
416                                                 sizeof(efi_runtime_services_t));
417         if (runtime != NULL) {
418                 /*
419                  * We will only need *early* access to the following
420                  * two EFI runtime services before set_virtual_address_map
421                  * is invoked.
422                  */
423                 efi_phys.get_time = (efi_get_time_t *) runtime->get_time;
424                 efi_phys.set_virtual_address_map =
425                         (efi_set_virtual_address_map_t *)
426                                 runtime->set_virtual_address_map;
427         } else
428                 printk(KERN_ERR PFX "Could not map the runtime service table!\n");
429
430         /* Map the EFI memory map for use until paging_init() */
431
432         memmap.map = (efi_memory_desc_t *)
433                 boot_ioremap((unsigned long) EFI_MEMMAP, EFI_MEMMAP_SIZE);
434
435         if (memmap.map == NULL)
436                 printk(KERN_ERR PFX "Could not map the EFI memory map!\n");
437
438         if (EFI_MEMDESC_SIZE != sizeof(efi_memory_desc_t)) {
439                 printk(KERN_WARNING PFX "Warning! Kernel-defined memdesc doesn't "
440                            "match the one from EFI!\n");
441         }
442 #if EFI_DEBUG
443         print_efi_memmap();
444 #endif
445 }
446
447 /*
448  * This function will switch the EFI runtime services to virtual mode.
449  * Essentially, look through the EFI memmap and map every region that
450  * has the runtime attribute bit set in its memory descriptor and update
451  * that memory descriptor with the virtual address obtained from ioremap().
452  * This enables the runtime services to be called without having to
453  * thunk back into physical mode for every invocation.
454  */
455
456 void __init efi_enter_virtual_mode(void)
457 {
458         efi_memory_desc_t *md;
459         efi_status_t status;
460         int i;
461
462         efi.systab = NULL;
463
464         for (i = 0; i < memmap.nr_map; i++) {
465                 md = &memmap.map[i];
466
467                 if (md->attribute & EFI_MEMORY_RUNTIME) {
468                         md->virt_addr =
469                                 (unsigned long)ioremap(md->phys_addr,
470                                         md->num_pages << EFI_PAGE_SHIFT);
471                         if (!(unsigned long)md->virt_addr) {
472                                 printk(KERN_ERR PFX "ioremap of 0x%lX failed\n",
473                                         (unsigned long)md->phys_addr);
474                         }
475
476                         if (((unsigned long)md->phys_addr <=
477                                         (unsigned long)efi_phys.systab) &&
478                                 ((unsigned long)efi_phys.systab <
479                                         md->phys_addr +
480                                         ((unsigned long)md->num_pages <<
481                                                 EFI_PAGE_SHIFT))) {
482                                 unsigned long addr;
483
484                                 addr = md->virt_addr - md->phys_addr +
485                                                 (unsigned long)efi_phys.systab;
486                                 efi.systab = (efi_system_table_t *)addr;
487                         }
488                 }
489         }
490
491         if (!efi.systab)
492                 BUG();
493
494         status = phys_efi_set_virtual_address_map(
495                         sizeof(efi_memory_desc_t) * memmap.nr_map,
496                         sizeof(efi_memory_desc_t),
497                         memmap.desc_version,
498                         memmap.phys_map);
499
500         if (status != EFI_SUCCESS) {
501                 printk (KERN_ALERT "You are screwed! "
502                         "Unable to switch EFI into virtual mode "
503                         "(status=%lx)\n", status);
504                 panic("EFI call to SetVirtualAddressMap() failed!");
505         }
506
507         /*
508          * Now that EFI is in virtual mode, update the function
509          * pointers in the runtime service table to the new virtual addresses.
510          */
511
512         efi.get_time = (efi_get_time_t *) efi.systab->runtime->get_time;
513         efi.set_time = (efi_set_time_t *) efi.systab->runtime->set_time;
514         efi.get_wakeup_time = (efi_get_wakeup_time_t *)
515                                         efi.systab->runtime->get_wakeup_time;
516         efi.set_wakeup_time = (efi_set_wakeup_time_t *)
517                                         efi.systab->runtime->set_wakeup_time;
518         efi.get_variable = (efi_get_variable_t *)
519                                         efi.systab->runtime->get_variable;
520         efi.get_next_variable = (efi_get_next_variable_t *)
521                                         efi.systab->runtime->get_next_variable;
522         efi.set_variable = (efi_set_variable_t *)
523                                         efi.systab->runtime->set_variable;
524         efi.get_next_high_mono_count = (efi_get_next_high_mono_count_t *)
525                                         efi.systab->runtime->get_next_high_mono_count;
526         efi.reset_system = (efi_reset_system_t *)
527                                         efi.systab->runtime->reset_system;
528 }
529
530 void __init
531 efi_initialize_iomem_resources(struct resource *code_resource,
532                                struct resource *data_resource)
533 {
534         struct resource *res;
535         efi_memory_desc_t *md;
536         int i;
537
538         for (i = 0; i < memmap.nr_map; i++) {
539                 md = &memmap.map[i];
540
541                 if ((md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >
542                     0x100000000ULL)
543                         continue;
544                 res = alloc_bootmem_low(sizeof(struct resource));
545                 switch (md->type) {
546                 case EFI_RESERVED_TYPE:
547                         res->name = "Reserved Memory";
548                         break;
549                 case EFI_LOADER_CODE:
550                         res->name = "Loader Code";
551                         break;
552                 case EFI_LOADER_DATA:
553                         res->name = "Loader Data";
554                         break;
555                 case EFI_BOOT_SERVICES_DATA:
556                         res->name = "BootServices Data";
557                         break;
558                 case EFI_BOOT_SERVICES_CODE:
559                         res->name = "BootServices Code";
560                         break;
561                 case EFI_RUNTIME_SERVICES_CODE:
562                         res->name = "Runtime Service Code";
563                         break;
564                 case EFI_RUNTIME_SERVICES_DATA:
565                         res->name = "Runtime Service Data";
566                         break;
567                 case EFI_CONVENTIONAL_MEMORY:
568                         res->name = "Conventional Memory";
569                         break;
570                 case EFI_UNUSABLE_MEMORY:
571                         res->name = "Unusable Memory";
572                         break;
573                 case EFI_ACPI_RECLAIM_MEMORY:
574                         res->name = "ACPI Reclaim";
575                         break;
576                 case EFI_ACPI_MEMORY_NVS:
577                         res->name = "ACPI NVS";
578                         break;
579                 case EFI_MEMORY_MAPPED_IO:
580                         res->name = "Memory Mapped IO";
581                         break;
582                 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
583                         res->name = "Memory Mapped IO Port Space";
584                         break;
585                 default:
586                         res->name = "Reserved";
587                         break;
588                 }
589                 res->start = md->phys_addr;
590                 res->end = res->start + ((md->num_pages << EFI_PAGE_SHIFT) - 1);
591                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
592                 if (request_resource(&iomem_resource, res) < 0)
593                         printk(KERN_ERR PFX "Failed to allocate res %s : 0x%lx-0x%lx\n",
594                                 res->name, res->start, res->end);
595                 /*
596                  * We don't know which region contains kernel data so we try
597                  * it repeatedly and let the resource manager test it.
598                  */
599                 if (md->type == EFI_CONVENTIONAL_MEMORY) {
600                         request_resource(res, code_resource);
601                         request_resource(res, data_resource);
602 #ifdef CONFIG_KEXEC
603                         request_resource(res, &crashk_res);
604 #endif
605                 }
606         }
607 }
608
609 /*
610  * Convenience functions to obtain memory types and attributes
611  */
612
613 u32 efi_mem_type(unsigned long phys_addr)
614 {
615         efi_memory_desc_t *md;
616         int i;
617
618         for (i = 0; i < memmap.nr_map; i++) {
619                 md = &memmap.map[i];
620                 if ((md->phys_addr <= phys_addr) && (phys_addr <
621                         (md->phys_addr + (md-> num_pages << EFI_PAGE_SHIFT)) ))
622                         return md->type;
623         }
624         return 0;
625 }
626
627 u64 efi_mem_attributes(unsigned long phys_addr)
628 {
629         efi_memory_desc_t *md;
630         int i;
631
632         for (i = 0; i < memmap.nr_map; i++) {
633                 md = &memmap.map[i];
634                 if ((md->phys_addr <= phys_addr) && (phys_addr <
635                         (md->phys_addr + (md-> num_pages << EFI_PAGE_SHIFT)) ))
636                         return md->attribute;
637         }
638         return 0;
639 }