Merge git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6
[pv_ops_mirror.git] / arch / ia64 / kernel / efi.c
blob728d7247a1a69688da32f12386d8fcb2165a1439
1 /*
2 * Extensible Firmware Interface
4 * Based on Extensible Firmware Interface Specification version 0.9
5 * April 30, 1999
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>
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
19 * Implemented EFI runtime services and virtual mode calls. --davidm
21 * Goutham Rao: <goutham.rao@intel.com>
22 * Skip non-WB memory and ignore empty memory ranges.
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>
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>
41 #define EFI_DEBUG 0
43 extern efi_status_t efi_call_phys (void *, ...);
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;
50 #define efi_call_virt(f, args...) (*(f))(args)
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; \
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; \
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; \
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; \
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; \
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; \
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) \
103 struct ia64_fpreg fr[6]; \
104 efi_time_t *atm = NULL; \
105 efi_status_t ret; \
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; \
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) \
122 struct ia64_fpreg fr[6]; \
123 u32 *aattr = NULL; \
124 efi_status_t ret; \
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; \
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) \
142 struct ia64_fpreg fr[6]; \
143 efi_status_t ret; \
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; \
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) \
159 struct ia64_fpreg fr[6]; \
160 efi_status_t ret; \
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; \
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) \
175 struct ia64_fpreg fr[6]; \
176 efi_status_t ret; \
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; \
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) \
191 struct ia64_fpreg fr[6]; \
192 efi_char16_t *adata = NULL; \
194 if (data) \
195 adata = adjust_arg(data); \
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); \
205 #define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
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)
217 #define id(arg) arg
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)
229 void
230 efi_gettimeofday (struct timespec *ts)
232 efi_time_t tm;
234 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
235 memset(ts, 0, sizeof(*ts));
236 return;
239 ts->tv_sec = mktime(tm.year, tm.month, tm.day,
240 tm.hour, tm.minute, tm.second);
241 ts->tv_nsec = tm.nanosecond;
244 static int
245 is_memory_available (efi_memory_desc_t *md)
247 if (!(md->attribute & EFI_MEMORY_WB))
248 return 0;
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;
258 return 0;
261 typedef struct kern_memdesc {
262 u64 attribute;
263 u64 start;
264 u64 num_pages;
265 } kern_memdesc_t;
267 static kern_memdesc_t *kern_memmap;
269 #define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
271 static inline u64
272 kmd_end(kern_memdesc_t *kmd)
274 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
277 static inline u64
278 efi_md_end(efi_memory_desc_t *md)
280 return (md->phys_addr + efi_md_size(md));
283 static inline int
284 efi_wb(efi_memory_desc_t *md)
286 return (md->attribute & EFI_MEMORY_WB);
289 static inline int
290 efi_uc(efi_memory_desc_t *md)
292 return (md->attribute & EFI_MEMORY_UC);
295 static void
296 walk (efi_freemem_callback_t callback, void *arg, u64 attr)
298 kern_memdesc_t *k;
299 u64 start, end, voff;
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;
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.
317 void
318 efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
320 walk(callback, arg, EFI_MEMORY_WB);
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.
327 void
328 efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
330 walk(callback, arg, EFI_MEMORY_UC);
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
338 void *
339 efi_get_pal_addr (void)
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;
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;
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;
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;
362 * The only ITLB entry in region 7 that is used is the one
363 * installed by __start(). That entry covers a 64MB range.
365 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
366 vaddr = PAGE_OFFSET + md->phys_addr;
369 * We must check that the PAL mapping won't overlap with the
370 * kernel mapping.
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.
381 if ((vaddr & mask) == (KERNEL_START & mask)) {
382 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
383 __func__);
384 continue;
387 if (efi_md_size(md) > IA64_GRANULE_SIZE)
388 panic("Whoa! PAL code size bigger than a granule!");
390 #if EFI_DEBUG
391 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
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);
401 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
402 __func__);
403 return NULL;
406 void
407 efi_map_pal_code (void)
409 void *pal_vaddr = efi_get_pal_addr ();
410 u64 psr;
412 if (!pal_vaddr)
413 return;
416 * Cannot write to CRx with PSR.ic=1
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 */
426 void __init
427 efi_init (void)
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;
437 * It's too early to be able to use the standard kernel command line
438 * support...
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;
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);
461 efi.systab = __va(ia64_boot_param->efi_systab);
464 * Verify the EFI Table
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);
476 config_tables = __va(efi.systab->tables);
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';
486 printk(KERN_INFO "EFI v%u.%.02u by %s:",
487 efi.systab->hdr.revision >> 16,
488 efi.systab->hdr.revision & 0xffff, vendor);
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;
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);
520 printk("\n");
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;
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;
537 #if EFI_DEBUG
538 /* print EFI memory map: */
540 efi_memory_desc_t *md;
541 void *p;
543 for (i = 0, p = efi_map_start; p < efi_map_end;
544 ++i, p += efi_desc_size)
546 const char *unit;
547 unsigned long size;
549 md = p;
550 size = md->num_pages << EFI_PAGE_SHIFT;
552 if ((size >> 40) > 0) {
553 size >>= 40;
554 unit = "TB";
555 } else if ((size >> 30) > 0) {
556 size >>= 30;
557 unit = "GB";
558 } else if ((size >> 20) > 0) {
559 size >>= 20;
560 unit = "MB";
561 } else {
562 size >>= 10;
563 unit = "KB";
566 printk("mem%02d: type=%2u, attr=0x%016lx, "
567 "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
568 i, md->type, md->attribute, md->phys_addr,
569 md->phys_addr + efi_md_size(md), size, unit);
572 #endif
574 efi_map_pal_code();
575 efi_enter_virtual_mode();
578 void
579 efi_enter_virtual_mode (void)
581 void *efi_map_start, *efi_map_end, *p;
582 efi_memory_desc_t *md;
583 efi_status_t status;
584 u64 efi_desc_size;
586 efi_map_start = __va(ia64_boot_param->efi_memmap);
587 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
588 efi_desc_size = ia64_boot_param->efi_memdesc_size;
590 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
591 md = p;
592 if (md->attribute & EFI_MEMORY_RUNTIME) {
594 * Some descriptors have multiple bits set, so the
595 * order of the tests is relevant.
597 if (md->attribute & EFI_MEMORY_WB) {
598 md->virt_addr = (u64) __va(md->phys_addr);
599 } else if (md->attribute & EFI_MEMORY_UC) {
600 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
601 } else if (md->attribute & EFI_MEMORY_WC) {
602 #if 0
603 md->virt_addr = ia64_remap(md->phys_addr,
604 (_PAGE_A |
605 _PAGE_P |
606 _PAGE_D |
607 _PAGE_MA_WC |
608 _PAGE_PL_0 |
609 _PAGE_AR_RW));
610 #else
611 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
612 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
613 #endif
614 } else if (md->attribute & EFI_MEMORY_WT) {
615 #if 0
616 md->virt_addr = ia64_remap(md->phys_addr,
617 (_PAGE_A |
618 _PAGE_P |
619 _PAGE_D |
620 _PAGE_MA_WT |
621 _PAGE_PL_0 |
622 _PAGE_AR_RW));
623 #else
624 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
625 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
626 #endif
631 status = efi_call_phys(__va(runtime->set_virtual_address_map),
632 ia64_boot_param->efi_memmap_size,
633 efi_desc_size,
634 ia64_boot_param->efi_memdesc_version,
635 ia64_boot_param->efi_memmap);
636 if (status != EFI_SUCCESS) {
637 printk(KERN_WARNING "warning: unable to switch EFI into "
638 "virtual mode (status=%lu)\n", status);
639 return;
643 * Now that EFI is in virtual mode, we call the EFI functions more
644 * efficiently:
646 efi.get_time = virt_get_time;
647 efi.set_time = virt_set_time;
648 efi.get_wakeup_time = virt_get_wakeup_time;
649 efi.set_wakeup_time = virt_set_wakeup_time;
650 efi.get_variable = virt_get_variable;
651 efi.get_next_variable = virt_get_next_variable;
652 efi.set_variable = virt_set_variable;
653 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
654 efi.reset_system = virt_reset_system;
658 * Walk the EFI memory map looking for the I/O port range. There can only be
659 * one entry of this type, other I/O port ranges should be described via ACPI.
662 efi_get_iobase (void)
664 void *efi_map_start, *efi_map_end, *p;
665 efi_memory_desc_t *md;
666 u64 efi_desc_size;
668 efi_map_start = __va(ia64_boot_param->efi_memmap);
669 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
670 efi_desc_size = ia64_boot_param->efi_memdesc_size;
672 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
673 md = p;
674 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
675 if (md->attribute & EFI_MEMORY_UC)
676 return md->phys_addr;
679 return 0;
682 static struct kern_memdesc *
683 kern_memory_descriptor (unsigned long phys_addr)
685 struct kern_memdesc *md;
687 for (md = kern_memmap; md->start != ~0UL; md++) {
688 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
689 return md;
691 return NULL;
694 static efi_memory_desc_t *
695 efi_memory_descriptor (unsigned long phys_addr)
697 void *efi_map_start, *efi_map_end, *p;
698 efi_memory_desc_t *md;
699 u64 efi_desc_size;
701 efi_map_start = __va(ia64_boot_param->efi_memmap);
702 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
703 efi_desc_size = ia64_boot_param->efi_memdesc_size;
705 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
706 md = p;
708 if (phys_addr - md->phys_addr < efi_md_size(md))
709 return md;
711 return NULL;
714 static int
715 efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
717 void *efi_map_start, *efi_map_end, *p;
718 efi_memory_desc_t *md;
719 u64 efi_desc_size;
720 unsigned long end;
722 efi_map_start = __va(ia64_boot_param->efi_memmap);
723 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
724 efi_desc_size = ia64_boot_param->efi_memdesc_size;
726 end = phys_addr + size;
728 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
729 md = p;
730 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
731 return 1;
733 return 0;
737 efi_mem_type (unsigned long phys_addr)
739 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
741 if (md)
742 return md->type;
743 return 0;
747 efi_mem_attributes (unsigned long phys_addr)
749 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
751 if (md)
752 return md->attribute;
753 return 0;
755 EXPORT_SYMBOL(efi_mem_attributes);
758 efi_mem_attribute (unsigned long phys_addr, unsigned long size)
760 unsigned long end = phys_addr + size;
761 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
762 u64 attr;
764 if (!md)
765 return 0;
768 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
769 * the kernel that firmware needs this region mapped.
771 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
772 do {
773 unsigned long md_end = efi_md_end(md);
775 if (end <= md_end)
776 return attr;
778 md = efi_memory_descriptor(md_end);
779 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
780 return 0;
781 } while (md);
782 return 0; /* never reached */
786 kern_mem_attribute (unsigned long phys_addr, unsigned long size)
788 unsigned long end = phys_addr + size;
789 struct kern_memdesc *md;
790 u64 attr;
793 * This is a hack for ioremap calls before we set up kern_memmap.
794 * Maybe we should do efi_memmap_init() earlier instead.
796 if (!kern_memmap) {
797 attr = efi_mem_attribute(phys_addr, size);
798 if (attr & EFI_MEMORY_WB)
799 return EFI_MEMORY_WB;
800 return 0;
803 md = kern_memory_descriptor(phys_addr);
804 if (!md)
805 return 0;
807 attr = md->attribute;
808 do {
809 unsigned long md_end = kmd_end(md);
811 if (end <= md_end)
812 return attr;
814 md = kern_memory_descriptor(md_end);
815 if (!md || md->attribute != attr)
816 return 0;
817 } while (md);
818 return 0; /* never reached */
820 EXPORT_SYMBOL(kern_mem_attribute);
823 valid_phys_addr_range (unsigned long phys_addr, unsigned long size)
825 u64 attr;
828 * /dev/mem reads and writes use copy_to_user(), which implicitly
829 * uses a granule-sized kernel identity mapping. It's really
830 * only safe to do this for regions in kern_memmap. For more
831 * details, see Documentation/ia64/aliasing.txt.
833 attr = kern_mem_attribute(phys_addr, size);
834 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
835 return 1;
836 return 0;
840 valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
842 unsigned long phys_addr = pfn << PAGE_SHIFT;
843 u64 attr;
845 attr = efi_mem_attribute(phys_addr, size);
848 * /dev/mem mmap uses normal user pages, so we don't need the entire
849 * granule, but the entire region we're mapping must support the same
850 * attribute.
852 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
853 return 1;
856 * Intel firmware doesn't tell us about all the MMIO regions, so
857 * in general we have to allow mmap requests. But if EFI *does*
858 * tell us about anything inside this region, we should deny it.
859 * The user can always map a smaller region to avoid the overlap.
861 if (efi_memmap_intersects(phys_addr, size))
862 return 0;
864 return 1;
867 pgprot_t
868 phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
869 pgprot_t vma_prot)
871 unsigned long phys_addr = pfn << PAGE_SHIFT;
872 u64 attr;
875 * For /dev/mem mmap, we use user mappings, but if the region is
876 * in kern_memmap (and hence may be covered by a kernel mapping),
877 * we must use the same attribute as the kernel mapping.
879 attr = kern_mem_attribute(phys_addr, size);
880 if (attr & EFI_MEMORY_WB)
881 return pgprot_cacheable(vma_prot);
882 else if (attr & EFI_MEMORY_UC)
883 return pgprot_noncached(vma_prot);
886 * Some chipsets don't support UC access to memory. If
887 * WB is supported, we prefer that.
889 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
890 return pgprot_cacheable(vma_prot);
892 return pgprot_noncached(vma_prot);
895 int __init
896 efi_uart_console_only(void)
898 efi_status_t status;
899 char *s, name[] = "ConOut";
900 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
901 efi_char16_t *utf16, name_utf16[32];
902 unsigned char data[1024];
903 unsigned long size = sizeof(data);
904 struct efi_generic_dev_path *hdr, *end_addr;
905 int uart = 0;
907 /* Convert to UTF-16 */
908 utf16 = name_utf16;
909 s = name;
910 while (*s)
911 *utf16++ = *s++ & 0x7f;
912 *utf16 = 0;
914 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
915 if (status != EFI_SUCCESS) {
916 printk(KERN_ERR "No EFI %s variable?\n", name);
917 return 0;
920 hdr = (struct efi_generic_dev_path *) data;
921 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
922 while (hdr < end_addr) {
923 if (hdr->type == EFI_DEV_MSG &&
924 hdr->sub_type == EFI_DEV_MSG_UART)
925 uart = 1;
926 else if (hdr->type == EFI_DEV_END_PATH ||
927 hdr->type == EFI_DEV_END_PATH2) {
928 if (!uart)
929 return 0;
930 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
931 return 1;
932 uart = 0;
934 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
936 printk(KERN_ERR "Malformed %s value\n", name);
937 return 0;
941 * Look for the first granule aligned memory descriptor memory
942 * that is big enough to hold EFI memory map. Make sure this
943 * descriptor is atleast granule sized so it does not get trimmed
945 struct kern_memdesc *
946 find_memmap_space (void)
948 u64 contig_low=0, contig_high=0;
949 u64 as = 0, ae;
950 void *efi_map_start, *efi_map_end, *p, *q;
951 efi_memory_desc_t *md, *pmd = NULL, *check_md;
952 u64 space_needed, efi_desc_size;
953 unsigned long total_mem = 0;
955 efi_map_start = __va(ia64_boot_param->efi_memmap);
956 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
957 efi_desc_size = ia64_boot_param->efi_memdesc_size;
960 * Worst case: we need 3 kernel descriptors for each efi descriptor
961 * (if every entry has a WB part in the middle, and UC head and tail),
962 * plus one for the end marker.
964 space_needed = sizeof(kern_memdesc_t) *
965 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
967 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
968 md = p;
969 if (!efi_wb(md)) {
970 continue;
972 if (pmd == NULL || !efi_wb(pmd) ||
973 efi_md_end(pmd) != md->phys_addr) {
974 contig_low = GRANULEROUNDUP(md->phys_addr);
975 contig_high = efi_md_end(md);
976 for (q = p + efi_desc_size; q < efi_map_end;
977 q += efi_desc_size) {
978 check_md = q;
979 if (!efi_wb(check_md))
980 break;
981 if (contig_high != check_md->phys_addr)
982 break;
983 contig_high = efi_md_end(check_md);
985 contig_high = GRANULEROUNDDOWN(contig_high);
987 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
988 continue;
990 /* Round ends inward to granule boundaries */
991 as = max(contig_low, md->phys_addr);
992 ae = min(contig_high, efi_md_end(md));
994 /* keep within max_addr= and min_addr= command line arg */
995 as = max(as, min_addr);
996 ae = min(ae, max_addr);
997 if (ae <= as)
998 continue;
1000 /* avoid going over mem= command line arg */
1001 if (total_mem + (ae - as) > mem_limit)
1002 ae -= total_mem + (ae - as) - mem_limit;
1004 if (ae <= as)
1005 continue;
1007 if (ae - as > space_needed)
1008 break;
1010 if (p >= efi_map_end)
1011 panic("Can't allocate space for kernel memory descriptors");
1013 return __va(as);
1017 * Walk the EFI memory map and gather all memory available for kernel
1018 * to use. We can allocate partial granules only if the unavailable
1019 * parts exist, and are WB.
1021 unsigned long
1022 efi_memmap_init(unsigned long *s, unsigned long *e)
1024 struct kern_memdesc *k, *prev = NULL;
1025 u64 contig_low=0, contig_high=0;
1026 u64 as, ae, lim;
1027 void *efi_map_start, *efi_map_end, *p, *q;
1028 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1029 u64 efi_desc_size;
1030 unsigned long total_mem = 0;
1032 k = kern_memmap = find_memmap_space();
1034 efi_map_start = __va(ia64_boot_param->efi_memmap);
1035 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1036 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1038 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1039 md = p;
1040 if (!efi_wb(md)) {
1041 if (efi_uc(md) &&
1042 (md->type == EFI_CONVENTIONAL_MEMORY ||
1043 md->type == EFI_BOOT_SERVICES_DATA)) {
1044 k->attribute = EFI_MEMORY_UC;
1045 k->start = md->phys_addr;
1046 k->num_pages = md->num_pages;
1047 k++;
1049 continue;
1051 if (pmd == NULL || !efi_wb(pmd) ||
1052 efi_md_end(pmd) != md->phys_addr) {
1053 contig_low = GRANULEROUNDUP(md->phys_addr);
1054 contig_high = efi_md_end(md);
1055 for (q = p + efi_desc_size; q < efi_map_end;
1056 q += efi_desc_size) {
1057 check_md = q;
1058 if (!efi_wb(check_md))
1059 break;
1060 if (contig_high != check_md->phys_addr)
1061 break;
1062 contig_high = efi_md_end(check_md);
1064 contig_high = GRANULEROUNDDOWN(contig_high);
1066 if (!is_memory_available(md))
1067 continue;
1069 #ifdef CONFIG_CRASH_DUMP
1070 /* saved_max_pfn should ignore max_addr= command line arg */
1071 if (saved_max_pfn < (efi_md_end(md) >> PAGE_SHIFT))
1072 saved_max_pfn = (efi_md_end(md) >> PAGE_SHIFT);
1073 #endif
1075 * Round ends inward to granule boundaries
1076 * Give trimmings to uncached allocator
1078 if (md->phys_addr < contig_low) {
1079 lim = min(efi_md_end(md), contig_low);
1080 if (efi_uc(md)) {
1081 if (k > kern_memmap &&
1082 (k-1)->attribute == EFI_MEMORY_UC &&
1083 kmd_end(k-1) == md->phys_addr) {
1084 (k-1)->num_pages +=
1085 (lim - md->phys_addr)
1086 >> EFI_PAGE_SHIFT;
1087 } else {
1088 k->attribute = EFI_MEMORY_UC;
1089 k->start = md->phys_addr;
1090 k->num_pages = (lim - md->phys_addr)
1091 >> EFI_PAGE_SHIFT;
1092 k++;
1095 as = contig_low;
1096 } else
1097 as = md->phys_addr;
1099 if (efi_md_end(md) > contig_high) {
1100 lim = max(md->phys_addr, contig_high);
1101 if (efi_uc(md)) {
1102 if (lim == md->phys_addr && k > kern_memmap &&
1103 (k-1)->attribute == EFI_MEMORY_UC &&
1104 kmd_end(k-1) == md->phys_addr) {
1105 (k-1)->num_pages += md->num_pages;
1106 } else {
1107 k->attribute = EFI_MEMORY_UC;
1108 k->start = lim;
1109 k->num_pages = (efi_md_end(md) - lim)
1110 >> EFI_PAGE_SHIFT;
1111 k++;
1114 ae = contig_high;
1115 } else
1116 ae = efi_md_end(md);
1118 /* keep within max_addr= and min_addr= command line arg */
1119 as = max(as, min_addr);
1120 ae = min(ae, max_addr);
1121 if (ae <= as)
1122 continue;
1124 /* avoid going over mem= command line arg */
1125 if (total_mem + (ae - as) > mem_limit)
1126 ae -= total_mem + (ae - as) - mem_limit;
1128 if (ae <= as)
1129 continue;
1130 if (prev && kmd_end(prev) == md->phys_addr) {
1131 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1132 total_mem += ae - as;
1133 continue;
1135 k->attribute = EFI_MEMORY_WB;
1136 k->start = as;
1137 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1138 total_mem += ae - as;
1139 prev = k++;
1141 k->start = ~0L; /* end-marker */
1143 /* reserve the memory we are using for kern_memmap */
1144 *s = (u64)kern_memmap;
1145 *e = (u64)++k;
1147 return total_mem;
1150 void
1151 efi_initialize_iomem_resources(struct resource *code_resource,
1152 struct resource *data_resource,
1153 struct resource *bss_resource)
1155 struct resource *res;
1156 void *efi_map_start, *efi_map_end, *p;
1157 efi_memory_desc_t *md;
1158 u64 efi_desc_size;
1159 char *name;
1160 unsigned long flags;
1162 efi_map_start = __va(ia64_boot_param->efi_memmap);
1163 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1164 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1166 res = NULL;
1168 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1169 md = p;
1171 if (md->num_pages == 0) /* should not happen */
1172 continue;
1174 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1175 switch (md->type) {
1177 case EFI_MEMORY_MAPPED_IO:
1178 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1179 continue;
1181 case EFI_LOADER_CODE:
1182 case EFI_LOADER_DATA:
1183 case EFI_BOOT_SERVICES_DATA:
1184 case EFI_BOOT_SERVICES_CODE:
1185 case EFI_CONVENTIONAL_MEMORY:
1186 if (md->attribute & EFI_MEMORY_WP) {
1187 name = "System ROM";
1188 flags |= IORESOURCE_READONLY;
1189 } else {
1190 name = "System RAM";
1192 break;
1194 case EFI_ACPI_MEMORY_NVS:
1195 name = "ACPI Non-volatile Storage";
1196 break;
1198 case EFI_UNUSABLE_MEMORY:
1199 name = "reserved";
1200 flags |= IORESOURCE_DISABLED;
1201 break;
1203 case EFI_RESERVED_TYPE:
1204 case EFI_RUNTIME_SERVICES_CODE:
1205 case EFI_RUNTIME_SERVICES_DATA:
1206 case EFI_ACPI_RECLAIM_MEMORY:
1207 default:
1208 name = "reserved";
1209 break;
1212 if ((res = kzalloc(sizeof(struct resource),
1213 GFP_KERNEL)) == NULL) {
1214 printk(KERN_ERR
1215 "failed to allocate resource for iomem\n");
1216 return;
1219 res->name = name;
1220 res->start = md->phys_addr;
1221 res->end = md->phys_addr + efi_md_size(md) - 1;
1222 res->flags = flags;
1224 if (insert_resource(&iomem_resource, res) < 0)
1225 kfree(res);
1226 else {
1228 * We don't know which region contains
1229 * kernel data so we try it repeatedly and
1230 * let the resource manager test it.
1232 insert_resource(res, code_resource);
1233 insert_resource(res, data_resource);
1234 insert_resource(res, bss_resource);
1235 #ifdef CONFIG_KEXEC
1236 insert_resource(res, &efi_memmap_res);
1237 insert_resource(res, &boot_param_res);
1238 if (crashk_res.end > crashk_res.start)
1239 insert_resource(res, &crashk_res);
1240 #endif
1245 #ifdef CONFIG_KEXEC
1246 /* find a block of memory aligned to 64M exclude reserved regions
1247 rsvd_regions are sorted
1249 unsigned long __init
1250 kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1252 int i;
1253 u64 start, end;
1254 u64 alignment = 1UL << _PAGE_SIZE_64M;
1255 void *efi_map_start, *efi_map_end, *p;
1256 efi_memory_desc_t *md;
1257 u64 efi_desc_size;
1259 efi_map_start = __va(ia64_boot_param->efi_memmap);
1260 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1261 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1263 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1264 md = p;
1265 if (!efi_wb(md))
1266 continue;
1267 start = ALIGN(md->phys_addr, alignment);
1268 end = efi_md_end(md);
1269 for (i = 0; i < n; i++) {
1270 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1271 if (__pa(r[i].start) > start + size)
1272 return start;
1273 start = ALIGN(__pa(r[i].end), alignment);
1274 if (i < n-1 &&
1275 __pa(r[i+1].start) < start + size)
1276 continue;
1277 else
1278 break;
1281 if (end > start + size)
1282 return start;
1285 printk(KERN_WARNING
1286 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1287 return ~0UL;
1289 #endif
1291 #ifdef CONFIG_PROC_VMCORE
1292 /* locate the size find a the descriptor at a certain address */
1293 unsigned long __init
1294 vmcore_find_descriptor_size (unsigned long address)
1296 void *efi_map_start, *efi_map_end, *p;
1297 efi_memory_desc_t *md;
1298 u64 efi_desc_size;
1299 unsigned long ret = 0;
1301 efi_map_start = __va(ia64_boot_param->efi_memmap);
1302 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1303 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1305 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1306 md = p;
1307 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1308 && md->phys_addr == address) {
1309 ret = efi_md_size(md);
1310 break;
1314 if (ret == 0)
1315 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1317 return ret;
1319 #endif