?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.25-rc4
[linux-2.6/next.git] / arch / ia64 / kernel / efi.c
blob919070a9aed7184c87ea43d83855963bb170d40e
1 /*
2 * Extensible Firmware Interface
3 *
4 * Based on Extensible Firmware Interface Specification version 0.9
5 * April 30, 1999
6 *
7 * Copyright (C) 1999 VA Linux Systems
8 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
9 * Copyright (C) 1999-2003 Hewlett-Packard Co.
10 * David Mosberger-Tang <davidm@hpl.hp.com>
11 * Stephane Eranian <eranian@hpl.hp.com>
12 * (c) Copyright 2006 Hewlett-Packard Development Company, L.P.
13 * Bjorn Helgaas <bjorn.helgaas@hp.com>
14 *
15 * All EFI Runtime Services are not implemented yet as EFI only
16 * supports physical mode addressing on SoftSDV. This is to be fixed
17 * in a future version. --drummond 1999-07-20
18 *
19 * Implemented EFI runtime services and virtual mode calls. --davidm
20 *
21 * Goutham Rao: <goutham.rao@intel.com>
22 * Skip non-WB memory and ignore empty memory ranges.
23 */
24 #include <linux/module.h>
25 #include <linux/bootmem.h>
26 #include <linux/kernel.h>
27 #include <linux/init.h>
28 #include <linux/types.h>
29 #include <linux/time.h>
30 #include <linux/efi.h>
31 #include <linux/kexec.h>
32 #include <linux/mm.h>
33
34 #include <asm/io.h>
35 #include <asm/kregs.h>
36 #include <asm/meminit.h>
37 #include <asm/pgtable.h>
38 #include <asm/processor.h>
39 #include <asm/mca.h>
40
41 #define EFI_DEBUG 0
42
43 extern efi_status_t efi_call_phys (void *, ...);
44
45 struct efi efi;
46 EXPORT_SYMBOL(efi);
47 static efi_runtime_services_t *runtime;
48 static unsigned long mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
49
50 #define efi_call_virt(f, args...) (*(f))(args)
51
52 #define STUB_GET_TIME(prefix, adjust_arg) \
53 static efi_status_t \
54 prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
55 { \
56 struct ia64_fpreg fr[6]; \
57 efi_time_cap_t *atc = NULL; \
58 efi_status_t ret; \
59 \
60 if (tc) \
61 atc = adjust_arg(tc); \
62 ia64_save_scratch_fpregs(fr); \
63 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \
64 adjust_arg(tm), atc); \
65 ia64_load_scratch_fpregs(fr); \
66 return ret; \
67 }
68
69 #define STUB_SET_TIME(prefix, adjust_arg) \
70 static efi_status_t \
71 prefix##_set_time (efi_time_t *tm) \
72 { \
73 struct ia64_fpreg fr[6]; \
74 efi_status_t ret; \
75 \
76 ia64_save_scratch_fpregs(fr); \
77 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
78 adjust_arg(tm)); \
79 ia64_load_scratch_fpregs(fr); \
80 return ret; \
81 }
82
83 #define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
84 static efi_status_t \
85 prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
86 efi_time_t *tm) \
87 { \
88 struct ia64_fpreg fr[6]; \
89 efi_status_t ret; \
90 \
91 ia64_save_scratch_fpregs(fr); \
92 ret = efi_call_##prefix( \
93 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
94 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
95 ia64_load_scratch_fpregs(fr); \
96 return ret; \
97 }
98
99 #define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
100 static efi_status_t \
101 prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
102 { \
103 struct ia64_fpreg fr[6]; \
104 efi_time_t *atm = NULL; \
105 efi_status_t ret; \
106 \
107 if (tm) \
108 atm = adjust_arg(tm); \
109 ia64_save_scratch_fpregs(fr); \
110 ret = efi_call_##prefix( \
111 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
112 enabled, atm); \
113 ia64_load_scratch_fpregs(fr); \
114 return ret; \
115 }
116
117 #define STUB_GET_VARIABLE(prefix, adjust_arg) \
118 static efi_status_t \
119 prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
120 unsigned long *data_size, void *data) \
121 { \
122 struct ia64_fpreg fr[6]; \
123 u32 *aattr = NULL; \
124 efi_status_t ret; \
125 \
126 if (attr) \
127 aattr = adjust_arg(attr); \
128 ia64_save_scratch_fpregs(fr); \
129 ret = efi_call_##prefix( \
130 (efi_get_variable_t *) __va(runtime->get_variable), \
131 adjust_arg(name), adjust_arg(vendor), aattr, \
132 adjust_arg(data_size), adjust_arg(data)); \
133 ia64_load_scratch_fpregs(fr); \
134 return ret; \
135 }
136
137 #define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
138 static efi_status_t \
139 prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
140 efi_guid_t *vendor) \
141 { \
142 struct ia64_fpreg fr[6]; \
143 efi_status_t ret; \
144 \
145 ia64_save_scratch_fpregs(fr); \
146 ret = efi_call_##prefix( \
147 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \
148 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
149 ia64_load_scratch_fpregs(fr); \
150 return ret; \
151 }
152
153 #define STUB_SET_VARIABLE(prefix, adjust_arg) \
154 static efi_status_t \
155 prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
156 unsigned long attr, unsigned long data_size, \
157 void *data) \
158 { \
159 struct ia64_fpreg fr[6]; \
160 efi_status_t ret; \
161 \
162 ia64_save_scratch_fpregs(fr); \
163 ret = efi_call_##prefix( \
164 (efi_set_variable_t *) __va(runtime->set_variable), \
165 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
166 adjust_arg(data)); \
167 ia64_load_scratch_fpregs(fr); \
168 return ret; \
169 }
170
171 #define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
172 static efi_status_t \
173 prefix##_get_next_high_mono_count (u32 *count) \
174 { \
175 struct ia64_fpreg fr[6]; \
176 efi_status_t ret; \
177 \
178 ia64_save_scratch_fpregs(fr); \
179 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
180 __va(runtime->get_next_high_mono_count), \
181 adjust_arg(count)); \
182 ia64_load_scratch_fpregs(fr); \
183 return ret; \
184 }
185
186 #define STUB_RESET_SYSTEM(prefix, adjust_arg) \
187 static void \
188 prefix##_reset_system (int reset_type, efi_status_t status, \
189 unsigned long data_size, efi_char16_t *data) \
190 { \
191 struct ia64_fpreg fr[6]; \
192 efi_char16_t *adata = NULL; \
193 \
194 if (data) \
195 adata = adjust_arg(data); \
196 \
197 ia64_save_scratch_fpregs(fr); \
198 efi_call_##prefix( \
199 (efi_reset_system_t *) __va(runtime->reset_system), \
200 reset_type, status, data_size, adata); \
201 /* should not return, but just in case... */ \
202 ia64_load_scratch_fpregs(fr); \
203 }
204
205 #define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
206
207 STUB_GET_TIME(phys, phys_ptr)
208 STUB_SET_TIME(phys, phys_ptr)
209 STUB_GET_WAKEUP_TIME(phys, phys_ptr)
210 STUB_SET_WAKEUP_TIME(phys, phys_ptr)
211 STUB_GET_VARIABLE(phys, phys_ptr)
212 STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
213 STUB_SET_VARIABLE(phys, phys_ptr)
214 STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
215 STUB_RESET_SYSTEM(phys, phys_ptr)
216
217 #define id(arg) arg
218
219 STUB_GET_TIME(virt, id)
220 STUB_SET_TIME(virt, id)
221 STUB_GET_WAKEUP_TIME(virt, id)
222 STUB_SET_WAKEUP_TIME(virt, id)
223 STUB_GET_VARIABLE(virt, id)
224 STUB_GET_NEXT_VARIABLE(virt, id)
225 STUB_SET_VARIABLE(virt, id)
226 STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
227 STUB_RESET_SYSTEM(virt, id)
228
229 void
230 efi_gettimeofday (struct timespec *ts)
231 {
232 efi_time_t tm;
233
234 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
235 memset(ts, 0, sizeof(*ts));
236 return;
237 }
238
239 ts->tv_sec = mktime(tm.year, tm.month, tm.day,
240 tm.hour, tm.minute, tm.second);
241 ts->tv_nsec = tm.nanosecond;
242 }
243
244 static int
245 is_memory_available (efi_memory_desc_t *md)
246 {
247 if (!(md->attribute & EFI_MEMORY_WB))
248 return 0;
249
250 switch (md->type) {
251 case EFI_LOADER_CODE:
252 case EFI_LOADER_DATA:
253 case EFI_BOOT_SERVICES_CODE:
254 case EFI_BOOT_SERVICES_DATA:
255 case EFI_CONVENTIONAL_MEMORY:
256 return 1;
257 }
258 return 0;
259 }
260
261 typedef struct kern_memdesc {
262 u64 attribute;
263 u64 start;
264 u64 num_pages;
265 } kern_memdesc_t;
266
267 static kern_memdesc_t *kern_memmap;
268
269 #define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
270
271 static inline u64
272 kmd_end(kern_memdesc_t *kmd)
273 {
274 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
275 }
276
277 static inline u64
278 efi_md_end(efi_memory_desc_t *md)
279 {
280 return (md->phys_addr + efi_md_size(md));
281 }
282
283 static inline int
284 efi_wb(efi_memory_desc_t *md)
285 {
286 return (md->attribute & EFI_MEMORY_WB);
287 }
288
289 static inline int
290 efi_uc(efi_memory_desc_t *md)
291 {
292 return (md->attribute & EFI_MEMORY_UC);
293 }
294
295 static void
296 walk (efi_freemem_callback_t callback, void *arg, u64 attr)
297 {
298 kern_memdesc_t *k;
299 u64 start, end, voff;
300
301 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
302 for (k = kern_memmap; k->start != ~0UL; k++) {
303 if (k->attribute != attr)
304 continue;
305 start = PAGE_ALIGN(k->start);
306 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
307 if (start < end)
308 if ((*callback)(start + voff, end + voff, arg) < 0)
309 return;
310 }
311 }
312
313 /*
314 * Walk the EFI memory map and call CALLBACK once for each EFI memory
315 * descriptor that has memory that is available for OS use.
316 */
317 void
318 efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
319 {
320 walk(callback, arg, EFI_MEMORY_WB);
321 }
322
323 /*
324 * Walk the EFI memory map and call CALLBACK once for each EFI memory
325 * descriptor that has memory that is available for uncached allocator.
326 */
327 void
328 efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
329 {
330 walk(callback, arg, EFI_MEMORY_UC);
331 }
332
333 /*
334 * Look for the PAL_CODE region reported by EFI and map it using an
335 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
336 * Abstraction Layer chapter 11 in ADAG
337 */
338 void *
339 efi_get_pal_addr (void)
340 {
341 void *efi_map_start, *efi_map_end, *p;
342 efi_memory_desc_t *md;
343 u64 efi_desc_size;
344 int pal_code_count = 0;
345 u64 vaddr, mask;
346
347 efi_map_start = __va(ia64_boot_param->efi_memmap);
348 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
349 efi_desc_size = ia64_boot_param->efi_memdesc_size;
350
351 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
352 md = p;
353 if (md->type != EFI_PAL_CODE)
354 continue;
355
356 if (++pal_code_count > 1) {
357 printk(KERN_ERR "Too many EFI Pal Code memory ranges, "
358 "dropped @ %lx\n", md->phys_addr);
359 continue;
360 }
361 /*
362 * The only ITLB entry in region 7 that is used is the one
363 * installed by __start(). That entry covers a 64MB range.
364 */
365 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
366 vaddr = PAGE_OFFSET + md->phys_addr;
367
368 /*
369 * We must check that the PAL mapping won't overlap with the
370 * kernel mapping.
371 *
372 * PAL code is guaranteed to be aligned on a power of 2 between
373 * 4k and 256KB and that only one ITR is needed to map it. This
374 * implies that the PAL code is always aligned on its size,
375 * i.e., the closest matching page size supported by the TLB.
376 * Therefore PAL code is guaranteed never to cross a 64MB unless
377 * it is bigger than 64MB (very unlikely!). So for now the
378 * following test is enough to determine whether or not we need
379 * a dedicated ITR for the PAL code.
380 */
381 if ((vaddr & mask) == (KERNEL_START & mask)) {
382 printk(KERN_INFO "%s: no need to install ITR for "
383 "PAL code\n", __FUNCTION__);
384 continue;
385 }
386
387 if (efi_md_size(md) > IA64_GRANULE_SIZE)
388 panic("Whoa! PAL code size bigger than a granule!");
389
390 #if EFI_DEBUG
391 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
392
393 printk(KERN_INFO "CPU %d: mapping PAL code "
394 "[0x%lx-0x%lx) into [0x%lx-0x%lx)\n",
395 smp_processor_id(), md->phys_addr,
396 md->phys_addr + efi_md_size(md),
397 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
398 #endif
399 return __va(md->phys_addr);
400 }
401 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
402 __FUNCTION__);
403 return NULL;
404 }
405
406 void
407 efi_map_pal_code (void)
408 {
409 void *pal_vaddr = efi_get_pal_addr ();
410 u64 psr;
411
412 if (!pal_vaddr)
413 return;
414
415 /*
416 * Cannot write to CRx with PSR.ic=1
417 */
418 psr = ia64_clear_ic();
419 ia64_itr(0x1, IA64_TR_PALCODE,
420 GRANULEROUNDDOWN((unsigned long) pal_vaddr),
421 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
422 IA64_GRANULE_SHIFT);
423 ia64_set_psr(psr); /* restore psr */
424 }
425
426 void __init
427 efi_init (void)
428 {
429 void *efi_map_start, *efi_map_end;
430 efi_config_table_t *config_tables;
431 efi_char16_t *c16;
432 u64 efi_desc_size;
433 char *cp, vendor[100] = "unknown";
434 int i;
435
436 /*
437 * It's too early to be able to use the standard kernel command line
438 * support...
439 */
440 for (cp = boot_command_line; *cp; ) {
441 if (memcmp(cp, "mem=", 4) == 0) {
442 mem_limit = memparse(cp + 4, &cp);
443 } else if (memcmp(cp, "max_addr=", 9) == 0) {
444 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
445 } else if (memcmp(cp, "min_addr=", 9) == 0) {
446 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
447 } else {
448 while (*cp != ' ' && *cp)
449 ++cp;
450 while (*cp == ' ')
451 ++cp;
452 }
453 }
454 if (min_addr != 0UL)
455 printk(KERN_INFO "Ignoring memory below %luMB\n",
456 min_addr >> 20);
457 if (max_addr != ~0UL)
458 printk(KERN_INFO "Ignoring memory above %luMB\n",
459 max_addr >> 20);
460
461 efi.systab = __va(ia64_boot_param->efi_systab);
462
463 /*
464 * Verify the EFI Table
465 */
466 if (efi.systab == NULL)
467 panic("Whoa! Can't find EFI system table.\n");
468 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
469 panic("Whoa! EFI system table signature incorrect\n");
470 if ((efi.systab->hdr.revision >> 16) == 0)
471 printk(KERN_WARNING "Warning: EFI system table version "
472 "%d.%02d, expected 1.00 or greater\n",
473 efi.systab->hdr.revision >> 16,
474 efi.systab->hdr.revision & 0xffff);
475
476 config_tables = __va(efi.systab->tables);
477
478 /* Show what we know for posterity */
479 c16 = __va(efi.systab->fw_vendor);
480 if (c16) {
481 for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i)
482 vendor[i] = *c16++;
483 vendor[i] = '\0';
484 }
485
486 printk(KERN_INFO "EFI v%u.%.02u by %s:",
487 efi.systab->hdr.revision >> 16,
488 efi.systab->hdr.revision & 0xffff, vendor);
489
490 efi.mps = EFI_INVALID_TABLE_ADDR;
491 efi.acpi = EFI_INVALID_TABLE_ADDR;
492 efi.acpi20 = EFI_INVALID_TABLE_ADDR;
493 efi.smbios = EFI_INVALID_TABLE_ADDR;
494 efi.sal_systab = EFI_INVALID_TABLE_ADDR;
495 efi.boot_info = EFI_INVALID_TABLE_ADDR;
496 efi.hcdp = EFI_INVALID_TABLE_ADDR;
497 efi.uga = EFI_INVALID_TABLE_ADDR;
498
499 for (i = 0; i < (int) efi.systab->nr_tables; i++) {
500 if (efi_guidcmp(config_tables[i].guid, MPS_TABLE_GUID) == 0) {
501 efi.mps = config_tables[i].table;
502 printk(" MPS=0x%lx", config_tables[i].table);
503 } else if (efi_guidcmp(config_tables[i].guid, ACPI_20_TABLE_GUID) == 0) {
504 efi.acpi20 = config_tables[i].table;
505 printk(" ACPI 2.0=0x%lx", config_tables[i].table);
506 } else if (efi_guidcmp(config_tables[i].guid, ACPI_TABLE_GUID) == 0) {
507 efi.acpi = config_tables[i].table;
508 printk(" ACPI=0x%lx", config_tables[i].table);
509 } else if (efi_guidcmp(config_tables[i].guid, SMBIOS_TABLE_GUID) == 0) {
510 efi.smbios = config_tables[i].table;
511 printk(" SMBIOS=0x%lx", config_tables[i].table);
512 } else if (efi_guidcmp(config_tables[i].guid, SAL_SYSTEM_TABLE_GUID) == 0) {
513 efi.sal_systab = config_tables[i].table;
514 printk(" SALsystab=0x%lx", config_tables[i].table);
515 } else if (efi_guidcmp(config_tables[i].guid, HCDP_TABLE_GUID) == 0) {
516 efi.hcdp = config_tables[i].table;
517 printk(" HCDP=0x%lx", config_tables[i].table);
518 }
519 }
520 printk("\n");
521
522 runtime = __va(efi.systab->runtime);
523 efi.get_time = phys_get_time;
524 efi.set_time = phys_set_time;
525 efi.get_wakeup_time = phys_get_wakeup_time;
526 efi.set_wakeup_time = phys_set_wakeup_time;
527 efi.get_variable = phys_get_variable;
528 efi.get_next_variable = phys_get_next_variable;
529 efi.set_variable = phys_set_variable;
530 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
531 efi.reset_system = phys_reset_system;
532
533 efi_map_start = __va(ia64_boot_param->efi_memmap);
534 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
535 efi_desc_size = ia64_boot_param->efi_memdesc_size;
536
537 #if EFI_DEBUG
538 /* print EFI memory map: */
539 {
540 efi_memory_desc_t *md;
541 void *p;
542
543 for (i = 0, p = efi_map_start; p < efi_map_end;
544 ++i, p += efi_desc_size)
545 {
546 md = p;
547 printk("mem%02u: type=%u, attr=0x%lx, "
548 "range=[0x%016lx-0x%016lx) (%luMB)\n",
549 i, md->type, md->attribute, md->phys_addr,
550 md->phys_addr + efi_md_size(md),
551 md->num_pages >> (20 - EFI_PAGE_SHIFT));
552 }
553 }
554 #endif
555
556 efi_map_pal_code();
557 efi_enter_virtual_mode();
558 }
559
560 void
561 efi_enter_virtual_mode (void)
562 {
563 void *efi_map_start, *efi_map_end, *p;
564 efi_memory_desc_t *md;
565 efi_status_t status;
566 u64 efi_desc_size;
567
568 efi_map_start = __va(ia64_boot_param->efi_memmap);
569 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
570 efi_desc_size = ia64_boot_param->efi_memdesc_size;
571
572 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
573 md = p;
574 if (md->attribute & EFI_MEMORY_RUNTIME) {
575 /*
576 * Some descriptors have multiple bits set, so the
577 * order of the tests is relevant.
578 */
579 if (md->attribute & EFI_MEMORY_WB) {
580 md->virt_addr = (u64) __va(md->phys_addr);
581 } else if (md->attribute & EFI_MEMORY_UC) {
582 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
583 } else if (md->attribute & EFI_MEMORY_WC) {
584 #if 0
585 md->virt_addr = ia64_remap(md->phys_addr,
586 (_PAGE_A |
587 _PAGE_P |
588 _PAGE_D |
589 _PAGE_MA_WC |
590 _PAGE_PL_0 |
591 _PAGE_AR_RW));
592 #else
593 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
594 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
595 #endif
596 } else if (md->attribute & EFI_MEMORY_WT) {
597 #if 0
598 md->virt_addr = ia64_remap(md->phys_addr,
599 (_PAGE_A |
600 _PAGE_P |
601 _PAGE_D |
602 _PAGE_MA_WT |
603 _PAGE_PL_0 |
604 _PAGE_AR_RW));
605 #else
606 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
607 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
608 #endif
609 }
610 }
611 }
612
613 status = efi_call_phys(__va(runtime->set_virtual_address_map),
614 ia64_boot_param->efi_memmap_size,
615 efi_desc_size,
616 ia64_boot_param->efi_memdesc_version,
617 ia64_boot_param->efi_memmap);
618 if (status != EFI_SUCCESS) {
619 printk(KERN_WARNING "warning: unable to switch EFI into "
620 "virtual mode (status=%lu)\n", status);
621 return;
622 }
623
624 /*
625 * Now that EFI is in virtual mode, we call the EFI functions more
626 * efficiently:
627 */
628 efi.get_time = virt_get_time;
629 efi.set_time = virt_set_time;
630 efi.get_wakeup_time = virt_get_wakeup_time;
631 efi.set_wakeup_time = virt_set_wakeup_time;
632 efi.get_variable = virt_get_variable;
633 efi.get_next_variable = virt_get_next_variable;
634 efi.set_variable = virt_set_variable;
635 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
636 efi.reset_system = virt_reset_system;
637 }
638
639 /*
640 * Walk the EFI memory map looking for the I/O port range. There can only be
641 * one entry of this type, other I/O port ranges should be described via ACPI.
642 */
643 u64
644 efi_get_iobase (void)
645 {
646 void *efi_map_start, *efi_map_end, *p;
647 efi_memory_desc_t *md;
648 u64 efi_desc_size;
649
650 efi_map_start = __va(ia64_boot_param->efi_memmap);
651 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
652 efi_desc_size = ia64_boot_param->efi_memdesc_size;
653
654 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
655 md = p;
656 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
657 if (md->attribute & EFI_MEMORY_UC)
658 return md->phys_addr;
659 }
660 }
661 return 0;
662 }
663
664 static struct kern_memdesc *
665 kern_memory_descriptor (unsigned long phys_addr)
666 {
667 struct kern_memdesc *md;
668
669 for (md = kern_memmap; md->start != ~0UL; md++) {
670 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
671 return md;
672 }
673 return NULL;
674 }
675
676 static efi_memory_desc_t *
677 efi_memory_descriptor (unsigned long phys_addr)
678 {
679 void *efi_map_start, *efi_map_end, *p;
680 efi_memory_desc_t *md;
681 u64 efi_desc_size;
682
683 efi_map_start = __va(ia64_boot_param->efi_memmap);
684 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
685 efi_desc_size = ia64_boot_param->efi_memdesc_size;
686
687 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
688 md = p;
689
690 if (phys_addr - md->phys_addr < efi_md_size(md))
691 return md;
692 }
693 return NULL;
694 }
695
696 static int
697 efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
698 {
699 void *efi_map_start, *efi_map_end, *p;
700 efi_memory_desc_t *md;
701 u64 efi_desc_size;
702 unsigned long end;
703
704 efi_map_start = __va(ia64_boot_param->efi_memmap);
705 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
706 efi_desc_size = ia64_boot_param->efi_memdesc_size;
707
708 end = phys_addr + size;
709
710 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
711 md = p;
712 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
713 return 1;
714 }
715 return 0;
716 }
717
718 u32
719 efi_mem_type (unsigned long phys_addr)
720 {
721 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
722
723 if (md)
724 return md->type;
725 return 0;
726 }
727
728 u64
729 efi_mem_attributes (unsigned long phys_addr)
730 {
731 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
732
733 if (md)
734 return md->attribute;
735 return 0;
736 }
737 EXPORT_SYMBOL(efi_mem_attributes);
738
739 u64
740 efi_mem_attribute (unsigned long phys_addr, unsigned long size)
741 {
742 unsigned long end = phys_addr + size;
743 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
744 u64 attr;
745
746 if (!md)
747 return 0;
748
749 /*
750 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
751 * the kernel that firmware needs this region mapped.
752 */
753 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
754 do {
755 unsigned long md_end = efi_md_end(md);
756
757 if (end <= md_end)
758 return attr;
759
760 md = efi_memory_descriptor(md_end);
761 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
762 return 0;
763 } while (md);
764 return 0; /* never reached */
765 }
766
767 u64
768 kern_mem_attribute (unsigned long phys_addr, unsigned long size)
769 {
770 unsigned long end = phys_addr + size;
771 struct kern_memdesc *md;
772 u64 attr;
773
774 /*
775 * This is a hack for ioremap calls before we set up kern_memmap.
776 * Maybe we should do efi_memmap_init() earlier instead.
777 */
778 if (!kern_memmap) {
779 attr = efi_mem_attribute(phys_addr, size);
780 if (attr & EFI_MEMORY_WB)
781 return EFI_MEMORY_WB;
782 return 0;
783 }
784
785 md = kern_memory_descriptor(phys_addr);
786 if (!md)
787 return 0;
788
789 attr = md->attribute;
790 do {
791 unsigned long md_end = kmd_end(md);
792
793 if (end <= md_end)
794 return attr;
795
796 md = kern_memory_descriptor(md_end);
797 if (!md || md->attribute != attr)
798 return 0;
799 } while (md);
800 return 0; /* never reached */
801 }
802 EXPORT_SYMBOL(kern_mem_attribute);
803
804 int
805 valid_phys_addr_range (unsigned long phys_addr, unsigned long size)
806 {
807 u64 attr;
808
809 /*
810 * /dev/mem reads and writes use copy_to_user(), which implicitly
811 * uses a granule-sized kernel identity mapping. It's really
812 * only safe to do this for regions in kern_memmap. For more
813 * details, see Documentation/ia64/aliasing.txt.
814 */
815 attr = kern_mem_attribute(phys_addr, size);
816 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
817 return 1;
818 return 0;
819 }
820
821 int
822 valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
823 {
824 unsigned long phys_addr = pfn << PAGE_SHIFT;
825 u64 attr;
826
827 attr = efi_mem_attribute(phys_addr, size);
828
829 /*
830 * /dev/mem mmap uses normal user pages, so we don't need the entire
831 * granule, but the entire region we're mapping must support the same
832 * attribute.
833 */
834 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
835 return 1;
836
837 /*
838 * Intel firmware doesn't tell us about all the MMIO regions, so
839 * in general we have to allow mmap requests. But if EFI *does*
840 * tell us about anything inside this region, we should deny it.
841 * The user can always map a smaller region to avoid the overlap.
842 */
843 if (efi_memmap_intersects(phys_addr, size))
844 return 0;
845
846 return 1;
847 }
848
849 pgprot_t
850 phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
851 pgprot_t vma_prot)
852 {
853 unsigned long phys_addr = pfn << PAGE_SHIFT;
854 u64 attr;
855
856 /*
857 * For /dev/mem mmap, we use user mappings, but if the region is
858 * in kern_memmap (and hence may be covered by a kernel mapping),
859 * we must use the same attribute as the kernel mapping.
860 */
861 attr = kern_mem_attribute(phys_addr, size);
862 if (attr & EFI_MEMORY_WB)
863 return pgprot_cacheable(vma_prot);
864 else if (attr & EFI_MEMORY_UC)
865 return pgprot_noncached(vma_prot);
866
867 /*
868 * Some chipsets don't support UC access to memory. If
869 * WB is supported, we prefer that.
870 */
871 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
872 return pgprot_cacheable(vma_prot);
873
874 return pgprot_noncached(vma_prot);
875 }
876
877 int __init
878 efi_uart_console_only(void)
879 {
880 efi_status_t status;
881 char *s, name[] = "ConOut";
882 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
883 efi_char16_t *utf16, name_utf16[32];
884 unsigned char data[1024];
885 unsigned long size = sizeof(data);
886 struct efi_generic_dev_path *hdr, *end_addr;
887 int uart = 0;
888
889 /* Convert to UTF-16 */
890 utf16 = name_utf16;
891 s = name;
892 while (*s)
893 *utf16++ = *s++ & 0x7f;
894 *utf16 = 0;
895
896 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
897 if (status != EFI_SUCCESS) {
898 printk(KERN_ERR "No EFI %s variable?\n", name);
899 return 0;
900 }
901
902 hdr = (struct efi_generic_dev_path *) data;
903 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
904 while (hdr < end_addr) {
905 if (hdr->type == EFI_DEV_MSG &&
906 hdr->sub_type == EFI_DEV_MSG_UART)
907 uart = 1;
908 else if (hdr->type == EFI_DEV_END_PATH ||
909 hdr->type == EFI_DEV_END_PATH2) {
910 if (!uart)
911 return 0;
912 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
913 return 1;
914 uart = 0;
915 }
916 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
917 }
918 printk(KERN_ERR "Malformed %s value\n", name);
919 return 0;
920 }
921
922 /*
923 * Look for the first granule aligned memory descriptor memory
924 * that is big enough to hold EFI memory map. Make sure this
925 * descriptor is atleast granule sized so it does not get trimmed
926 */
927 struct kern_memdesc *
928 find_memmap_space (void)
929 {
930 u64 contig_low=0, contig_high=0;
931 u64 as = 0, ae;
932 void *efi_map_start, *efi_map_end, *p, *q;
933 efi_memory_desc_t *md, *pmd = NULL, *check_md;
934 u64 space_needed, efi_desc_size;
935 unsigned long total_mem = 0;
936
937 efi_map_start = __va(ia64_boot_param->efi_memmap);
938 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
939 efi_desc_size = ia64_boot_param->efi_memdesc_size;
940
941 /*
942 * Worst case: we need 3 kernel descriptors for each efi descriptor
943 * (if every entry has a WB part in the middle, and UC head and tail),
944 * plus one for the end marker.
945 */
946 space_needed = sizeof(kern_memdesc_t) *
947 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
948
949 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
950 md = p;
951 if (!efi_wb(md)) {
952 continue;
953 }
954 if (pmd == NULL || !efi_wb(pmd) ||
955 efi_md_end(pmd) != md->phys_addr) {
956 contig_low = GRANULEROUNDUP(md->phys_addr);
957 contig_high = efi_md_end(md);
958 for (q = p + efi_desc_size; q < efi_map_end;
959 q += efi_desc_size) {
960 check_md = q;
961 if (!efi_wb(check_md))
962 break;
963 if (contig_high != check_md->phys_addr)
964 break;
965 contig_high = efi_md_end(check_md);
966 }
967 contig_high = GRANULEROUNDDOWN(contig_high);
968 }
969 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
970 continue;
971
972 /* Round ends inward to granule boundaries */
973 as = max(contig_low, md->phys_addr);
974 ae = min(contig_high, efi_md_end(md));
975
976 /* keep within max_addr= and min_addr= command line arg */
977 as = max(as, min_addr);
978 ae = min(ae, max_addr);
979 if (ae <= as)
980 continue;
981
982 /* avoid going over mem= command line arg */
983 if (total_mem + (ae - as) > mem_limit)
984 ae -= total_mem + (ae - as) - mem_limit;
985
986 if (ae <= as)
987 continue;
988
989 if (ae - as > space_needed)
990 break;
991 }
992 if (p >= efi_map_end)
993 panic("Can't allocate space for kernel memory descriptors");
994
995 return __va(as);
996 }
997
998 /*
999 * Walk the EFI memory map and gather all memory available for kernel
1000 * to use. We can allocate partial granules only if the unavailable
1001 * parts exist, and are WB.
1002 */
1003 unsigned long
1004 efi_memmap_init(unsigned long *s, unsigned long *e)
1005 {
1006 struct kern_memdesc *k, *prev = NULL;
1007 u64 contig_low=0, contig_high=0;
1008 u64 as, ae, lim;
1009 void *efi_map_start, *efi_map_end, *p, *q;
1010 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1011 u64 efi_desc_size;
1012 unsigned long total_mem = 0;
1013
1014 k = kern_memmap = find_memmap_space();
1015
1016 efi_map_start = __va(ia64_boot_param->efi_memmap);
1017 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1018 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1019
1020 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1021 md = p;
1022 if (!efi_wb(md)) {
1023 if (efi_uc(md) &&
1024 (md->type == EFI_CONVENTIONAL_MEMORY ||
1025 md->type == EFI_BOOT_SERVICES_DATA)) {
1026 k->attribute = EFI_MEMORY_UC;
1027 k->start = md->phys_addr;
1028 k->num_pages = md->num_pages;
1029 k++;
1030 }
1031 continue;
1032 }
1033 if (pmd == NULL || !efi_wb(pmd) ||
1034 efi_md_end(pmd) != md->phys_addr) {
1035 contig_low = GRANULEROUNDUP(md->phys_addr);
1036 contig_high = efi_md_end(md);
1037 for (q = p + efi_desc_size; q < efi_map_end;
1038 q += efi_desc_size) {
1039 check_md = q;
1040 if (!efi_wb(check_md))
1041 break;
1042 if (contig_high != check_md->phys_addr)
1043 break;
1044 contig_high = efi_md_end(check_md);
1045 }
1046 contig_high = GRANULEROUNDDOWN(contig_high);
1047 }
1048 if (!is_memory_available(md))
1049 continue;
1050
1051 #ifdef CONFIG_CRASH_DUMP
1052 /* saved_max_pfn should ignore max_addr= command line arg */
1053 if (saved_max_pfn < (efi_md_end(md) >> PAGE_SHIFT))
1054 saved_max_pfn = (efi_md_end(md) >> PAGE_SHIFT);
1055 #endif
1056 /*
1057 * Round ends inward to granule boundaries
1058 * Give trimmings to uncached allocator
1059 */
1060 if (md->phys_addr < contig_low) {
1061 lim = min(efi_md_end(md), contig_low);
1062 if (efi_uc(md)) {
1063 if (k > kern_memmap &&
1064 (k-1)->attribute == EFI_MEMORY_UC &&
1065 kmd_end(k-1) == md->phys_addr) {
1066 (k-1)->num_pages +=
1067 (lim - md->phys_addr)
1068 >> EFI_PAGE_SHIFT;
1069 } else {
1070 k->attribute = EFI_MEMORY_UC;
1071 k->start = md->phys_addr;
1072 k->num_pages = (lim - md->phys_addr)
1073 >> EFI_PAGE_SHIFT;
1074 k++;
1075 }
1076 }
1077 as = contig_low;
1078 } else
1079 as = md->phys_addr;
1080
1081 if (efi_md_end(md) > contig_high) {
1082 lim = max(md->phys_addr, contig_high);
1083 if (efi_uc(md)) {
1084 if (lim == md->phys_addr && k > kern_memmap &&
1085 (k-1)->attribute == EFI_MEMORY_UC &&
1086 kmd_end(k-1) == md->phys_addr) {
1087 (k-1)->num_pages += md->num_pages;
1088 } else {
1089 k->attribute = EFI_MEMORY_UC;
1090 k->start = lim;
1091 k->num_pages = (efi_md_end(md) - lim)
1092 >> EFI_PAGE_SHIFT;
1093 k++;
1094 }
1095 }
1096 ae = contig_high;
1097 } else
1098 ae = efi_md_end(md);
1099
1100 /* keep within max_addr= and min_addr= command line arg */
1101 as = max(as, min_addr);
1102 ae = min(ae, max_addr);
1103 if (ae <= as)
1104 continue;
1105
1106 /* avoid going over mem= command line arg */
1107 if (total_mem + (ae - as) > mem_limit)
1108 ae -= total_mem + (ae - as) - mem_limit;
1109
1110 if (ae <= as)
1111 continue;
1112 if (prev && kmd_end(prev) == md->phys_addr) {
1113 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1114 total_mem += ae - as;
1115 continue;
1116 }
1117 k->attribute = EFI_MEMORY_WB;
1118 k->start = as;
1119 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1120 total_mem += ae - as;
1121 prev = k++;
1122 }
1123 k->start = ~0L; /* end-marker */
1124
1125 /* reserve the memory we are using for kern_memmap */
1126 *s = (u64)kern_memmap;
1127 *e = (u64)++k;
1128
1129 return total_mem;
1130 }
1131
1132 void
1133 efi_initialize_iomem_resources(struct resource *code_resource,
1134 struct resource *data_resource,
1135 struct resource *bss_resource)
1136 {
1137 struct resource *res;
1138 void *efi_map_start, *efi_map_end, *p;
1139 efi_memory_desc_t *md;
1140 u64 efi_desc_size;
1141 char *name;
1142 unsigned long flags;
1143
1144 efi_map_start = __va(ia64_boot_param->efi_memmap);
1145 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1146 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1147
1148 res = NULL;
1149
1150 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1151 md = p;
1152
1153 if (md->num_pages == 0) /* should not happen */
1154 continue;
1155
1156 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1157 switch (md->type) {
1158
1159 case EFI_MEMORY_MAPPED_IO:
1160 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1161 continue;
1162
1163 case EFI_LOADER_CODE:
1164 case EFI_LOADER_DATA:
1165 case EFI_BOOT_SERVICES_DATA:
1166 case EFI_BOOT_SERVICES_CODE:
1167 case EFI_CONVENTIONAL_MEMORY:
1168 if (md->attribute & EFI_MEMORY_WP) {
1169 name = "System ROM";
1170 flags |= IORESOURCE_READONLY;
1171 } else {
1172 name = "System RAM";
1173 }
1174 break;
1175
1176 case EFI_ACPI_MEMORY_NVS:
1177 name = "ACPI Non-volatile Storage";
1178 break;
1179
1180 case EFI_UNUSABLE_MEMORY:
1181 name = "reserved";
1182 flags |= IORESOURCE_DISABLED;
1183 break;
1184
1185 case EFI_RESERVED_TYPE:
1186 case EFI_RUNTIME_SERVICES_CODE:
1187 case EFI_RUNTIME_SERVICES_DATA:
1188 case EFI_ACPI_RECLAIM_MEMORY:
1189 default:
1190 name = "reserved";
1191 break;
1192 }
1193
1194 if ((res = kzalloc(sizeof(struct resource),
1195 GFP_KERNEL)) == NULL) {
1196 printk(KERN_ERR
1197 "failed to allocate resource for iomem\n");
1198 return;
1199 }
1200
1201 res->name = name;
1202 res->start = md->phys_addr;
1203 res->end = md->phys_addr + efi_md_size(md) - 1;
1204 res->flags = flags;
1205
1206 if (insert_resource(&iomem_resource, res) < 0)
1207 kfree(res);
1208 else {
1209 /*
1210 * We don't know which region contains
1211 * kernel data so we try it repeatedly and
1212 * let the resource manager test it.
1213 */
1214 insert_resource(res, code_resource);
1215 insert_resource(res, data_resource);
1216 insert_resource(res, bss_resource);
1217 #ifdef CONFIG_KEXEC
1218 insert_resource(res, &efi_memmap_res);
1219 insert_resource(res, &boot_param_res);
1220 if (crashk_res.end > crashk_res.start)
1221 insert_resource(res, &crashk_res);
1222 #endif
1223 }
1224 }
1225 }
1226
1227 #ifdef CONFIG_KEXEC
1228 /* find a block of memory aligned to 64M exclude reserved regions
1229 rsvd_regions are sorted
1230 */
1231 unsigned long __init
1232 kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1233 {
1234 int i;
1235 u64 start, end;
1236 u64 alignment = 1UL << _PAGE_SIZE_64M;
1237 void *efi_map_start, *efi_map_end, *p;
1238 efi_memory_desc_t *md;
1239 u64 efi_desc_size;
1240
1241 efi_map_start = __va(ia64_boot_param->efi_memmap);
1242 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1243 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1244
1245 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1246 md = p;
1247 if (!efi_wb(md))
1248 continue;
1249 start = ALIGN(md->phys_addr, alignment);
1250 end = efi_md_end(md);
1251 for (i = 0; i < n; i++) {
1252 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1253 if (__pa(r[i].start) > start + size)
1254 return start;
1255 start = ALIGN(__pa(r[i].end), alignment);
1256 if (i < n-1 &&
1257 __pa(r[i+1].start) < start + size)
1258 continue;
1259 else
1260 break;
1261 }
1262 }
1263 if (end > start + size)
1264 return start;
1265 }
1266
1267 printk(KERN_WARNING
1268 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1269 return ~0UL;
1270 }
1271 #endif
1272
1273 #ifdef CONFIG_PROC_VMCORE
1274 /* locate the size find a the descriptor at a certain address */
1275 unsigned long __init
1276 vmcore_find_descriptor_size (unsigned long address)
1277 {
1278 void *efi_map_start, *efi_map_end, *p;
1279 efi_memory_desc_t *md;
1280 u64 efi_desc_size;
1281 unsigned long ret = 0;
1282
1283 efi_map_start = __va(ia64_boot_param->efi_memmap);
1284 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1285 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1286
1287 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1288 md = p;
1289 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1290 && md->phys_addr == address) {
1291 ret = efi_md_size(md);
1292 break;
1293 }
1294 }
1295
1296 if (ret == 0)
1297 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1298
1299 return ret;
1300 }
1301 #endif
linux-next