ia64/pv_ops/binary patch: define paravirt_dv_serialize_data() and suppress false...
[linux-2.6/mini2440.git] / arch / ia64 / kernel / efi.c
blob7ef80e8161ce3dbad06d522ef5634e9e8520be3e
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>
40 #include <asm/tlbflush.h>
42 #define EFI_DEBUG 0
44 extern efi_status_t efi_call_phys (void *, ...);
46 struct efi efi;
47 EXPORT_SYMBOL(efi);
48 static efi_runtime_services_t *runtime;
49 static unsigned long mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
51 #define efi_call_virt(f, args...) (*(f))(args)
53 #define STUB_GET_TIME(prefix, adjust_arg) \
54 static efi_status_t \
55 prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
56 { \
57 struct ia64_fpreg fr[6]; \
58 efi_time_cap_t *atc = NULL; \
59 efi_status_t ret; \
61 if (tc) \
62 atc = adjust_arg(tc); \
63 ia64_save_scratch_fpregs(fr); \
64 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \
65 adjust_arg(tm), atc); \
66 ia64_load_scratch_fpregs(fr); \
67 return ret; \
70 #define STUB_SET_TIME(prefix, adjust_arg) \
71 static efi_status_t \
72 prefix##_set_time (efi_time_t *tm) \
73 { \
74 struct ia64_fpreg fr[6]; \
75 efi_status_t ret; \
77 ia64_save_scratch_fpregs(fr); \
78 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
79 adjust_arg(tm)); \
80 ia64_load_scratch_fpregs(fr); \
81 return ret; \
84 #define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
85 static efi_status_t \
86 prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
87 efi_time_t *tm) \
88 { \
89 struct ia64_fpreg fr[6]; \
90 efi_status_t ret; \
92 ia64_save_scratch_fpregs(fr); \
93 ret = efi_call_##prefix( \
94 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
95 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
96 ia64_load_scratch_fpregs(fr); \
97 return ret; \
100 #define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
101 static efi_status_t \
102 prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
104 struct ia64_fpreg fr[6]; \
105 efi_time_t *atm = NULL; \
106 efi_status_t ret; \
108 if (tm) \
109 atm = adjust_arg(tm); \
110 ia64_save_scratch_fpregs(fr); \
111 ret = efi_call_##prefix( \
112 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
113 enabled, atm); \
114 ia64_load_scratch_fpregs(fr); \
115 return ret; \
118 #define STUB_GET_VARIABLE(prefix, adjust_arg) \
119 static efi_status_t \
120 prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
121 unsigned long *data_size, void *data) \
123 struct ia64_fpreg fr[6]; \
124 u32 *aattr = NULL; \
125 efi_status_t ret; \
127 if (attr) \
128 aattr = adjust_arg(attr); \
129 ia64_save_scratch_fpregs(fr); \
130 ret = efi_call_##prefix( \
131 (efi_get_variable_t *) __va(runtime->get_variable), \
132 adjust_arg(name), adjust_arg(vendor), aattr, \
133 adjust_arg(data_size), adjust_arg(data)); \
134 ia64_load_scratch_fpregs(fr); \
135 return ret; \
138 #define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
139 static efi_status_t \
140 prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
141 efi_guid_t *vendor) \
143 struct ia64_fpreg fr[6]; \
144 efi_status_t ret; \
146 ia64_save_scratch_fpregs(fr); \
147 ret = efi_call_##prefix( \
148 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \
149 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
150 ia64_load_scratch_fpregs(fr); \
151 return ret; \
154 #define STUB_SET_VARIABLE(prefix, adjust_arg) \
155 static efi_status_t \
156 prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
157 unsigned long attr, unsigned long data_size, \
158 void *data) \
160 struct ia64_fpreg fr[6]; \
161 efi_status_t ret; \
163 ia64_save_scratch_fpregs(fr); \
164 ret = efi_call_##prefix( \
165 (efi_set_variable_t *) __va(runtime->set_variable), \
166 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
167 adjust_arg(data)); \
168 ia64_load_scratch_fpregs(fr); \
169 return ret; \
172 #define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
173 static efi_status_t \
174 prefix##_get_next_high_mono_count (u32 *count) \
176 struct ia64_fpreg fr[6]; \
177 efi_status_t ret; \
179 ia64_save_scratch_fpregs(fr); \
180 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
181 __va(runtime->get_next_high_mono_count), \
182 adjust_arg(count)); \
183 ia64_load_scratch_fpregs(fr); \
184 return ret; \
187 #define STUB_RESET_SYSTEM(prefix, adjust_arg) \
188 static void \
189 prefix##_reset_system (int reset_type, efi_status_t status, \
190 unsigned long data_size, efi_char16_t *data) \
192 struct ia64_fpreg fr[6]; \
193 efi_char16_t *adata = NULL; \
195 if (data) \
196 adata = adjust_arg(data); \
198 ia64_save_scratch_fpregs(fr); \
199 efi_call_##prefix( \
200 (efi_reset_system_t *) __va(runtime->reset_system), \
201 reset_type, status, data_size, adata); \
202 /* should not return, but just in case... */ \
203 ia64_load_scratch_fpregs(fr); \
206 #define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
208 STUB_GET_TIME(phys, phys_ptr)
209 STUB_SET_TIME(phys, phys_ptr)
210 STUB_GET_WAKEUP_TIME(phys, phys_ptr)
211 STUB_SET_WAKEUP_TIME(phys, phys_ptr)
212 STUB_GET_VARIABLE(phys, phys_ptr)
213 STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
214 STUB_SET_VARIABLE(phys, phys_ptr)
215 STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
216 STUB_RESET_SYSTEM(phys, phys_ptr)
218 #define id(arg) arg
220 STUB_GET_TIME(virt, id)
221 STUB_SET_TIME(virt, id)
222 STUB_GET_WAKEUP_TIME(virt, id)
223 STUB_SET_WAKEUP_TIME(virt, id)
224 STUB_GET_VARIABLE(virt, id)
225 STUB_GET_NEXT_VARIABLE(virt, id)
226 STUB_SET_VARIABLE(virt, id)
227 STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
228 STUB_RESET_SYSTEM(virt, id)
230 void
231 efi_gettimeofday (struct timespec *ts)
233 efi_time_t tm;
235 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
236 memset(ts, 0, sizeof(*ts));
237 return;
240 ts->tv_sec = mktime(tm.year, tm.month, tm.day,
241 tm.hour, tm.minute, tm.second);
242 ts->tv_nsec = tm.nanosecond;
245 static int
246 is_memory_available (efi_memory_desc_t *md)
248 if (!(md->attribute & EFI_MEMORY_WB))
249 return 0;
251 switch (md->type) {
252 case EFI_LOADER_CODE:
253 case EFI_LOADER_DATA:
254 case EFI_BOOT_SERVICES_CODE:
255 case EFI_BOOT_SERVICES_DATA:
256 case EFI_CONVENTIONAL_MEMORY:
257 return 1;
259 return 0;
262 typedef struct kern_memdesc {
263 u64 attribute;
264 u64 start;
265 u64 num_pages;
266 } kern_memdesc_t;
268 static kern_memdesc_t *kern_memmap;
270 #define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
272 static inline u64
273 kmd_end(kern_memdesc_t *kmd)
275 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
278 static inline u64
279 efi_md_end(efi_memory_desc_t *md)
281 return (md->phys_addr + efi_md_size(md));
284 static inline int
285 efi_wb(efi_memory_desc_t *md)
287 return (md->attribute & EFI_MEMORY_WB);
290 static inline int
291 efi_uc(efi_memory_desc_t *md)
293 return (md->attribute & EFI_MEMORY_UC);
296 static void
297 walk (efi_freemem_callback_t callback, void *arg, u64 attr)
299 kern_memdesc_t *k;
300 u64 start, end, voff;
302 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
303 for (k = kern_memmap; k->start != ~0UL; k++) {
304 if (k->attribute != attr)
305 continue;
306 start = PAGE_ALIGN(k->start);
307 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
308 if (start < end)
309 if ((*callback)(start + voff, end + voff, arg) < 0)
310 return;
315 * Walk the EFI memory map and call CALLBACK once for each EFI memory
316 * descriptor that has memory that is available for OS use.
318 void
319 efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
321 walk(callback, arg, EFI_MEMORY_WB);
325 * Walk the EFI memory map and call CALLBACK once for each EFI memory
326 * descriptor that has memory that is available for uncached allocator.
328 void
329 efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
331 walk(callback, arg, EFI_MEMORY_UC);
335 * Look for the PAL_CODE region reported by EFI and map it using an
336 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
337 * Abstraction Layer chapter 11 in ADAG
339 void *
340 efi_get_pal_addr (void)
342 void *efi_map_start, *efi_map_end, *p;
343 efi_memory_desc_t *md;
344 u64 efi_desc_size;
345 int pal_code_count = 0;
346 u64 vaddr, mask;
348 efi_map_start = __va(ia64_boot_param->efi_memmap);
349 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
350 efi_desc_size = ia64_boot_param->efi_memdesc_size;
352 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
353 md = p;
354 if (md->type != EFI_PAL_CODE)
355 continue;
357 if (++pal_code_count > 1) {
358 printk(KERN_ERR "Too many EFI Pal Code memory ranges, "
359 "dropped @ %lx\n", md->phys_addr);
360 continue;
363 * The only ITLB entry in region 7 that is used is the one
364 * installed by __start(). That entry covers a 64MB range.
366 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
367 vaddr = PAGE_OFFSET + md->phys_addr;
370 * We must check that the PAL mapping won't overlap with the
371 * kernel mapping.
373 * PAL code is guaranteed to be aligned on a power of 2 between
374 * 4k and 256KB and that only one ITR is needed to map it. This
375 * implies that the PAL code is always aligned on its size,
376 * i.e., the closest matching page size supported by the TLB.
377 * Therefore PAL code is guaranteed never to cross a 64MB unless
378 * it is bigger than 64MB (very unlikely!). So for now the
379 * following test is enough to determine whether or not we need
380 * a dedicated ITR for the PAL code.
382 if ((vaddr & mask) == (KERNEL_START & mask)) {
383 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
384 __func__);
385 continue;
388 if (efi_md_size(md) > IA64_GRANULE_SIZE)
389 panic("Whoa! PAL code size bigger than a granule!");
391 #if EFI_DEBUG
392 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
394 printk(KERN_INFO "CPU %d: mapping PAL code "
395 "[0x%lx-0x%lx) into [0x%lx-0x%lx)\n",
396 smp_processor_id(), md->phys_addr,
397 md->phys_addr + efi_md_size(md),
398 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
399 #endif
400 return __va(md->phys_addr);
402 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
403 __func__);
404 return NULL;
408 static u8 __init palo_checksum(u8 *buffer, u32 length)
410 u8 sum = 0;
411 u8 *end = buffer + length;
413 while (buffer < end)
414 sum = (u8) (sum + *(buffer++));
416 return sum;
420 * Parse and handle PALO table which is published at:
421 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
423 static void __init handle_palo(unsigned long palo_phys)
425 struct palo_table *palo = __va(palo_phys);
426 u8 checksum;
428 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
429 printk(KERN_INFO "PALO signature incorrect.\n");
430 return;
433 checksum = palo_checksum((u8 *)palo, palo->length);
434 if (checksum) {
435 printk(KERN_INFO "PALO checksum incorrect.\n");
436 return;
439 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
442 void
443 efi_map_pal_code (void)
445 void *pal_vaddr = efi_get_pal_addr ();
446 u64 psr;
448 if (!pal_vaddr)
449 return;
452 * Cannot write to CRx with PSR.ic=1
454 psr = ia64_clear_ic();
455 ia64_itr(0x1, IA64_TR_PALCODE,
456 GRANULEROUNDDOWN((unsigned long) pal_vaddr),
457 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
458 IA64_GRANULE_SHIFT);
459 paravirt_dv_serialize_data();
460 ia64_set_psr(psr); /* restore psr */
463 void __init
464 efi_init (void)
466 void *efi_map_start, *efi_map_end;
467 efi_config_table_t *config_tables;
468 efi_char16_t *c16;
469 u64 efi_desc_size;
470 char *cp, vendor[100] = "unknown";
471 int i;
472 unsigned long palo_phys;
475 * It's too early to be able to use the standard kernel command line
476 * support...
478 for (cp = boot_command_line; *cp; ) {
479 if (memcmp(cp, "mem=", 4) == 0) {
480 mem_limit = memparse(cp + 4, &cp);
481 } else if (memcmp(cp, "max_addr=", 9) == 0) {
482 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
483 } else if (memcmp(cp, "min_addr=", 9) == 0) {
484 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
485 } else {
486 while (*cp != ' ' && *cp)
487 ++cp;
488 while (*cp == ' ')
489 ++cp;
492 if (min_addr != 0UL)
493 printk(KERN_INFO "Ignoring memory below %luMB\n",
494 min_addr >> 20);
495 if (max_addr != ~0UL)
496 printk(KERN_INFO "Ignoring memory above %luMB\n",
497 max_addr >> 20);
499 efi.systab = __va(ia64_boot_param->efi_systab);
502 * Verify the EFI Table
504 if (efi.systab == NULL)
505 panic("Whoa! Can't find EFI system table.\n");
506 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
507 panic("Whoa! EFI system table signature incorrect\n");
508 if ((efi.systab->hdr.revision >> 16) == 0)
509 printk(KERN_WARNING "Warning: EFI system table version "
510 "%d.%02d, expected 1.00 or greater\n",
511 efi.systab->hdr.revision >> 16,
512 efi.systab->hdr.revision & 0xffff);
514 config_tables = __va(efi.systab->tables);
516 /* Show what we know for posterity */
517 c16 = __va(efi.systab->fw_vendor);
518 if (c16) {
519 for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i)
520 vendor[i] = *c16++;
521 vendor[i] = '\0';
524 printk(KERN_INFO "EFI v%u.%.02u by %s:",
525 efi.systab->hdr.revision >> 16,
526 efi.systab->hdr.revision & 0xffff, vendor);
528 efi.mps = EFI_INVALID_TABLE_ADDR;
529 efi.acpi = EFI_INVALID_TABLE_ADDR;
530 efi.acpi20 = EFI_INVALID_TABLE_ADDR;
531 efi.smbios = EFI_INVALID_TABLE_ADDR;
532 efi.sal_systab = EFI_INVALID_TABLE_ADDR;
533 efi.boot_info = EFI_INVALID_TABLE_ADDR;
534 efi.hcdp = EFI_INVALID_TABLE_ADDR;
535 efi.uga = EFI_INVALID_TABLE_ADDR;
537 palo_phys = EFI_INVALID_TABLE_ADDR;
539 for (i = 0; i < (int) efi.systab->nr_tables; i++) {
540 if (efi_guidcmp(config_tables[i].guid, MPS_TABLE_GUID) == 0) {
541 efi.mps = config_tables[i].table;
542 printk(" MPS=0x%lx", config_tables[i].table);
543 } else if (efi_guidcmp(config_tables[i].guid, ACPI_20_TABLE_GUID) == 0) {
544 efi.acpi20 = config_tables[i].table;
545 printk(" ACPI 2.0=0x%lx", config_tables[i].table);
546 } else if (efi_guidcmp(config_tables[i].guid, ACPI_TABLE_GUID) == 0) {
547 efi.acpi = config_tables[i].table;
548 printk(" ACPI=0x%lx", config_tables[i].table);
549 } else if (efi_guidcmp(config_tables[i].guid, SMBIOS_TABLE_GUID) == 0) {
550 efi.smbios = config_tables[i].table;
551 printk(" SMBIOS=0x%lx", config_tables[i].table);
552 } else if (efi_guidcmp(config_tables[i].guid, SAL_SYSTEM_TABLE_GUID) == 0) {
553 efi.sal_systab = config_tables[i].table;
554 printk(" SALsystab=0x%lx", config_tables[i].table);
555 } else if (efi_guidcmp(config_tables[i].guid, HCDP_TABLE_GUID) == 0) {
556 efi.hcdp = config_tables[i].table;
557 printk(" HCDP=0x%lx", config_tables[i].table);
558 } else if (efi_guidcmp(config_tables[i].guid,
559 PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID) == 0) {
560 palo_phys = config_tables[i].table;
561 printk(" PALO=0x%lx", config_tables[i].table);
564 printk("\n");
566 if (palo_phys != EFI_INVALID_TABLE_ADDR)
567 handle_palo(palo_phys);
569 runtime = __va(efi.systab->runtime);
570 efi.get_time = phys_get_time;
571 efi.set_time = phys_set_time;
572 efi.get_wakeup_time = phys_get_wakeup_time;
573 efi.set_wakeup_time = phys_set_wakeup_time;
574 efi.get_variable = phys_get_variable;
575 efi.get_next_variable = phys_get_next_variable;
576 efi.set_variable = phys_set_variable;
577 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
578 efi.reset_system = phys_reset_system;
580 efi_map_start = __va(ia64_boot_param->efi_memmap);
581 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
582 efi_desc_size = ia64_boot_param->efi_memdesc_size;
584 #if EFI_DEBUG
585 /* print EFI memory map: */
587 efi_memory_desc_t *md;
588 void *p;
590 for (i = 0, p = efi_map_start; p < efi_map_end;
591 ++i, p += efi_desc_size)
593 const char *unit;
594 unsigned long size;
596 md = p;
597 size = md->num_pages << EFI_PAGE_SHIFT;
599 if ((size >> 40) > 0) {
600 size >>= 40;
601 unit = "TB";
602 } else if ((size >> 30) > 0) {
603 size >>= 30;
604 unit = "GB";
605 } else if ((size >> 20) > 0) {
606 size >>= 20;
607 unit = "MB";
608 } else {
609 size >>= 10;
610 unit = "KB";
613 printk("mem%02d: type=%2u, attr=0x%016lx, "
614 "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
615 i, md->type, md->attribute, md->phys_addr,
616 md->phys_addr + efi_md_size(md), size, unit);
619 #endif
621 efi_map_pal_code();
622 efi_enter_virtual_mode();
625 void
626 efi_enter_virtual_mode (void)
628 void *efi_map_start, *efi_map_end, *p;
629 efi_memory_desc_t *md;
630 efi_status_t status;
631 u64 efi_desc_size;
633 efi_map_start = __va(ia64_boot_param->efi_memmap);
634 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
635 efi_desc_size = ia64_boot_param->efi_memdesc_size;
637 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
638 md = p;
639 if (md->attribute & EFI_MEMORY_RUNTIME) {
641 * Some descriptors have multiple bits set, so the
642 * order of the tests is relevant.
644 if (md->attribute & EFI_MEMORY_WB) {
645 md->virt_addr = (u64) __va(md->phys_addr);
646 } else if (md->attribute & EFI_MEMORY_UC) {
647 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
648 } else if (md->attribute & EFI_MEMORY_WC) {
649 #if 0
650 md->virt_addr = ia64_remap(md->phys_addr,
651 (_PAGE_A |
652 _PAGE_P |
653 _PAGE_D |
654 _PAGE_MA_WC |
655 _PAGE_PL_0 |
656 _PAGE_AR_RW));
657 #else
658 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
659 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
660 #endif
661 } else if (md->attribute & EFI_MEMORY_WT) {
662 #if 0
663 md->virt_addr = ia64_remap(md->phys_addr,
664 (_PAGE_A |
665 _PAGE_P |
666 _PAGE_D |
667 _PAGE_MA_WT |
668 _PAGE_PL_0 |
669 _PAGE_AR_RW));
670 #else
671 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
672 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
673 #endif
678 status = efi_call_phys(__va(runtime->set_virtual_address_map),
679 ia64_boot_param->efi_memmap_size,
680 efi_desc_size,
681 ia64_boot_param->efi_memdesc_version,
682 ia64_boot_param->efi_memmap);
683 if (status != EFI_SUCCESS) {
684 printk(KERN_WARNING "warning: unable to switch EFI into "
685 "virtual mode (status=%lu)\n", status);
686 return;
690 * Now that EFI is in virtual mode, we call the EFI functions more
691 * efficiently:
693 efi.get_time = virt_get_time;
694 efi.set_time = virt_set_time;
695 efi.get_wakeup_time = virt_get_wakeup_time;
696 efi.set_wakeup_time = virt_set_wakeup_time;
697 efi.get_variable = virt_get_variable;
698 efi.get_next_variable = virt_get_next_variable;
699 efi.set_variable = virt_set_variable;
700 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
701 efi.reset_system = virt_reset_system;
705 * Walk the EFI memory map looking for the I/O port range. There can only be
706 * one entry of this type, other I/O port ranges should be described via ACPI.
709 efi_get_iobase (void)
711 void *efi_map_start, *efi_map_end, *p;
712 efi_memory_desc_t *md;
713 u64 efi_desc_size;
715 efi_map_start = __va(ia64_boot_param->efi_memmap);
716 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
717 efi_desc_size = ia64_boot_param->efi_memdesc_size;
719 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
720 md = p;
721 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
722 if (md->attribute & EFI_MEMORY_UC)
723 return md->phys_addr;
726 return 0;
729 static struct kern_memdesc *
730 kern_memory_descriptor (unsigned long phys_addr)
732 struct kern_memdesc *md;
734 for (md = kern_memmap; md->start != ~0UL; md++) {
735 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
736 return md;
738 return NULL;
741 static efi_memory_desc_t *
742 efi_memory_descriptor (unsigned long phys_addr)
744 void *efi_map_start, *efi_map_end, *p;
745 efi_memory_desc_t *md;
746 u64 efi_desc_size;
748 efi_map_start = __va(ia64_boot_param->efi_memmap);
749 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
750 efi_desc_size = ia64_boot_param->efi_memdesc_size;
752 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
753 md = p;
755 if (phys_addr - md->phys_addr < efi_md_size(md))
756 return md;
758 return NULL;
761 static int
762 efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
764 void *efi_map_start, *efi_map_end, *p;
765 efi_memory_desc_t *md;
766 u64 efi_desc_size;
767 unsigned long end;
769 efi_map_start = __va(ia64_boot_param->efi_memmap);
770 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
771 efi_desc_size = ia64_boot_param->efi_memdesc_size;
773 end = phys_addr + size;
775 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
776 md = p;
777 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
778 return 1;
780 return 0;
784 efi_mem_type (unsigned long phys_addr)
786 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
788 if (md)
789 return md->type;
790 return 0;
794 efi_mem_attributes (unsigned long phys_addr)
796 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
798 if (md)
799 return md->attribute;
800 return 0;
802 EXPORT_SYMBOL(efi_mem_attributes);
805 efi_mem_attribute (unsigned long phys_addr, unsigned long size)
807 unsigned long end = phys_addr + size;
808 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
809 u64 attr;
811 if (!md)
812 return 0;
815 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
816 * the kernel that firmware needs this region mapped.
818 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
819 do {
820 unsigned long md_end = efi_md_end(md);
822 if (end <= md_end)
823 return attr;
825 md = efi_memory_descriptor(md_end);
826 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
827 return 0;
828 } while (md);
829 return 0; /* never reached */
833 kern_mem_attribute (unsigned long phys_addr, unsigned long size)
835 unsigned long end = phys_addr + size;
836 struct kern_memdesc *md;
837 u64 attr;
840 * This is a hack for ioremap calls before we set up kern_memmap.
841 * Maybe we should do efi_memmap_init() earlier instead.
843 if (!kern_memmap) {
844 attr = efi_mem_attribute(phys_addr, size);
845 if (attr & EFI_MEMORY_WB)
846 return EFI_MEMORY_WB;
847 return 0;
850 md = kern_memory_descriptor(phys_addr);
851 if (!md)
852 return 0;
854 attr = md->attribute;
855 do {
856 unsigned long md_end = kmd_end(md);
858 if (end <= md_end)
859 return attr;
861 md = kern_memory_descriptor(md_end);
862 if (!md || md->attribute != attr)
863 return 0;
864 } while (md);
865 return 0; /* never reached */
867 EXPORT_SYMBOL(kern_mem_attribute);
870 valid_phys_addr_range (unsigned long phys_addr, unsigned long size)
872 u64 attr;
875 * /dev/mem reads and writes use copy_to_user(), which implicitly
876 * uses a granule-sized kernel identity mapping. It's really
877 * only safe to do this for regions in kern_memmap. For more
878 * details, see Documentation/ia64/aliasing.txt.
880 attr = kern_mem_attribute(phys_addr, size);
881 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
882 return 1;
883 return 0;
887 valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
889 unsigned long phys_addr = pfn << PAGE_SHIFT;
890 u64 attr;
892 attr = efi_mem_attribute(phys_addr, size);
895 * /dev/mem mmap uses normal user pages, so we don't need the entire
896 * granule, but the entire region we're mapping must support the same
897 * attribute.
899 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
900 return 1;
903 * Intel firmware doesn't tell us about all the MMIO regions, so
904 * in general we have to allow mmap requests. But if EFI *does*
905 * tell us about anything inside this region, we should deny it.
906 * The user can always map a smaller region to avoid the overlap.
908 if (efi_memmap_intersects(phys_addr, size))
909 return 0;
911 return 1;
914 pgprot_t
915 phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
916 pgprot_t vma_prot)
918 unsigned long phys_addr = pfn << PAGE_SHIFT;
919 u64 attr;
922 * For /dev/mem mmap, we use user mappings, but if the region is
923 * in kern_memmap (and hence may be covered by a kernel mapping),
924 * we must use the same attribute as the kernel mapping.
926 attr = kern_mem_attribute(phys_addr, size);
927 if (attr & EFI_MEMORY_WB)
928 return pgprot_cacheable(vma_prot);
929 else if (attr & EFI_MEMORY_UC)
930 return pgprot_noncached(vma_prot);
933 * Some chipsets don't support UC access to memory. If
934 * WB is supported, we prefer that.
936 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
937 return pgprot_cacheable(vma_prot);
939 return pgprot_noncached(vma_prot);
942 int __init
943 efi_uart_console_only(void)
945 efi_status_t status;
946 char *s, name[] = "ConOut";
947 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
948 efi_char16_t *utf16, name_utf16[32];
949 unsigned char data[1024];
950 unsigned long size = sizeof(data);
951 struct efi_generic_dev_path *hdr, *end_addr;
952 int uart = 0;
954 /* Convert to UTF-16 */
955 utf16 = name_utf16;
956 s = name;
957 while (*s)
958 *utf16++ = *s++ & 0x7f;
959 *utf16 = 0;
961 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
962 if (status != EFI_SUCCESS) {
963 printk(KERN_ERR "No EFI %s variable?\n", name);
964 return 0;
967 hdr = (struct efi_generic_dev_path *) data;
968 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
969 while (hdr < end_addr) {
970 if (hdr->type == EFI_DEV_MSG &&
971 hdr->sub_type == EFI_DEV_MSG_UART)
972 uart = 1;
973 else if (hdr->type == EFI_DEV_END_PATH ||
974 hdr->type == EFI_DEV_END_PATH2) {
975 if (!uart)
976 return 0;
977 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
978 return 1;
979 uart = 0;
981 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
983 printk(KERN_ERR "Malformed %s value\n", name);
984 return 0;
988 * Look for the first granule aligned memory descriptor memory
989 * that is big enough to hold EFI memory map. Make sure this
990 * descriptor is atleast granule sized so it does not get trimmed
992 struct kern_memdesc *
993 find_memmap_space (void)
995 u64 contig_low=0, contig_high=0;
996 u64 as = 0, ae;
997 void *efi_map_start, *efi_map_end, *p, *q;
998 efi_memory_desc_t *md, *pmd = NULL, *check_md;
999 u64 space_needed, efi_desc_size;
1000 unsigned long total_mem = 0;
1002 efi_map_start = __va(ia64_boot_param->efi_memmap);
1003 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1004 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1007 * Worst case: we need 3 kernel descriptors for each efi descriptor
1008 * (if every entry has a WB part in the middle, and UC head and tail),
1009 * plus one for the end marker.
1011 space_needed = sizeof(kern_memdesc_t) *
1012 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
1014 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1015 md = p;
1016 if (!efi_wb(md)) {
1017 continue;
1019 if (pmd == NULL || !efi_wb(pmd) ||
1020 efi_md_end(pmd) != md->phys_addr) {
1021 contig_low = GRANULEROUNDUP(md->phys_addr);
1022 contig_high = efi_md_end(md);
1023 for (q = p + efi_desc_size; q < efi_map_end;
1024 q += efi_desc_size) {
1025 check_md = q;
1026 if (!efi_wb(check_md))
1027 break;
1028 if (contig_high != check_md->phys_addr)
1029 break;
1030 contig_high = efi_md_end(check_md);
1032 contig_high = GRANULEROUNDDOWN(contig_high);
1034 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
1035 continue;
1037 /* Round ends inward to granule boundaries */
1038 as = max(contig_low, md->phys_addr);
1039 ae = min(contig_high, efi_md_end(md));
1041 /* keep within max_addr= and min_addr= command line arg */
1042 as = max(as, min_addr);
1043 ae = min(ae, max_addr);
1044 if (ae <= as)
1045 continue;
1047 /* avoid going over mem= command line arg */
1048 if (total_mem + (ae - as) > mem_limit)
1049 ae -= total_mem + (ae - as) - mem_limit;
1051 if (ae <= as)
1052 continue;
1054 if (ae - as > space_needed)
1055 break;
1057 if (p >= efi_map_end)
1058 panic("Can't allocate space for kernel memory descriptors");
1060 return __va(as);
1064 * Walk the EFI memory map and gather all memory available for kernel
1065 * to use. We can allocate partial granules only if the unavailable
1066 * parts exist, and are WB.
1068 unsigned long
1069 efi_memmap_init(unsigned long *s, unsigned long *e)
1071 struct kern_memdesc *k, *prev = NULL;
1072 u64 contig_low=0, contig_high=0;
1073 u64 as, ae, lim;
1074 void *efi_map_start, *efi_map_end, *p, *q;
1075 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1076 u64 efi_desc_size;
1077 unsigned long total_mem = 0;
1079 k = kern_memmap = find_memmap_space();
1081 efi_map_start = __va(ia64_boot_param->efi_memmap);
1082 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1083 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1085 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1086 md = p;
1087 if (!efi_wb(md)) {
1088 if (efi_uc(md) &&
1089 (md->type == EFI_CONVENTIONAL_MEMORY ||
1090 md->type == EFI_BOOT_SERVICES_DATA)) {
1091 k->attribute = EFI_MEMORY_UC;
1092 k->start = md->phys_addr;
1093 k->num_pages = md->num_pages;
1094 k++;
1096 continue;
1098 if (pmd == NULL || !efi_wb(pmd) ||
1099 efi_md_end(pmd) != md->phys_addr) {
1100 contig_low = GRANULEROUNDUP(md->phys_addr);
1101 contig_high = efi_md_end(md);
1102 for (q = p + efi_desc_size; q < efi_map_end;
1103 q += efi_desc_size) {
1104 check_md = q;
1105 if (!efi_wb(check_md))
1106 break;
1107 if (contig_high != check_md->phys_addr)
1108 break;
1109 contig_high = efi_md_end(check_md);
1111 contig_high = GRANULEROUNDDOWN(contig_high);
1113 if (!is_memory_available(md))
1114 continue;
1116 #ifdef CONFIG_CRASH_DUMP
1117 /* saved_max_pfn should ignore max_addr= command line arg */
1118 if (saved_max_pfn < (efi_md_end(md) >> PAGE_SHIFT))
1119 saved_max_pfn = (efi_md_end(md) >> PAGE_SHIFT);
1120 #endif
1122 * Round ends inward to granule boundaries
1123 * Give trimmings to uncached allocator
1125 if (md->phys_addr < contig_low) {
1126 lim = min(efi_md_end(md), contig_low);
1127 if (efi_uc(md)) {
1128 if (k > kern_memmap &&
1129 (k-1)->attribute == EFI_MEMORY_UC &&
1130 kmd_end(k-1) == md->phys_addr) {
1131 (k-1)->num_pages +=
1132 (lim - md->phys_addr)
1133 >> EFI_PAGE_SHIFT;
1134 } else {
1135 k->attribute = EFI_MEMORY_UC;
1136 k->start = md->phys_addr;
1137 k->num_pages = (lim - md->phys_addr)
1138 >> EFI_PAGE_SHIFT;
1139 k++;
1142 as = contig_low;
1143 } else
1144 as = md->phys_addr;
1146 if (efi_md_end(md) > contig_high) {
1147 lim = max(md->phys_addr, contig_high);
1148 if (efi_uc(md)) {
1149 if (lim == md->phys_addr && k > kern_memmap &&
1150 (k-1)->attribute == EFI_MEMORY_UC &&
1151 kmd_end(k-1) == md->phys_addr) {
1152 (k-1)->num_pages += md->num_pages;
1153 } else {
1154 k->attribute = EFI_MEMORY_UC;
1155 k->start = lim;
1156 k->num_pages = (efi_md_end(md) - lim)
1157 >> EFI_PAGE_SHIFT;
1158 k++;
1161 ae = contig_high;
1162 } else
1163 ae = efi_md_end(md);
1165 /* keep within max_addr= and min_addr= command line arg */
1166 as = max(as, min_addr);
1167 ae = min(ae, max_addr);
1168 if (ae <= as)
1169 continue;
1171 /* avoid going over mem= command line arg */
1172 if (total_mem + (ae - as) > mem_limit)
1173 ae -= total_mem + (ae - as) - mem_limit;
1175 if (ae <= as)
1176 continue;
1177 if (prev && kmd_end(prev) == md->phys_addr) {
1178 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1179 total_mem += ae - as;
1180 continue;
1182 k->attribute = EFI_MEMORY_WB;
1183 k->start = as;
1184 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1185 total_mem += ae - as;
1186 prev = k++;
1188 k->start = ~0L; /* end-marker */
1190 /* reserve the memory we are using for kern_memmap */
1191 *s = (u64)kern_memmap;
1192 *e = (u64)++k;
1194 return total_mem;
1197 void
1198 efi_initialize_iomem_resources(struct resource *code_resource,
1199 struct resource *data_resource,
1200 struct resource *bss_resource)
1202 struct resource *res;
1203 void *efi_map_start, *efi_map_end, *p;
1204 efi_memory_desc_t *md;
1205 u64 efi_desc_size;
1206 char *name;
1207 unsigned long flags;
1209 efi_map_start = __va(ia64_boot_param->efi_memmap);
1210 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1211 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1213 res = NULL;
1215 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1216 md = p;
1218 if (md->num_pages == 0) /* should not happen */
1219 continue;
1221 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1222 switch (md->type) {
1224 case EFI_MEMORY_MAPPED_IO:
1225 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1226 continue;
1228 case EFI_LOADER_CODE:
1229 case EFI_LOADER_DATA:
1230 case EFI_BOOT_SERVICES_DATA:
1231 case EFI_BOOT_SERVICES_CODE:
1232 case EFI_CONVENTIONAL_MEMORY:
1233 if (md->attribute & EFI_MEMORY_WP) {
1234 name = "System ROM";
1235 flags |= IORESOURCE_READONLY;
1236 } else if (md->attribute == EFI_MEMORY_UC)
1237 name = "Uncached RAM";
1238 else
1239 name = "System RAM";
1240 break;
1242 case EFI_ACPI_MEMORY_NVS:
1243 name = "ACPI Non-volatile Storage";
1244 break;
1246 case EFI_UNUSABLE_MEMORY:
1247 name = "reserved";
1248 flags |= IORESOURCE_DISABLED;
1249 break;
1251 case EFI_RESERVED_TYPE:
1252 case EFI_RUNTIME_SERVICES_CODE:
1253 case EFI_RUNTIME_SERVICES_DATA:
1254 case EFI_ACPI_RECLAIM_MEMORY:
1255 default:
1256 name = "reserved";
1257 break;
1260 if ((res = kzalloc(sizeof(struct resource),
1261 GFP_KERNEL)) == NULL) {
1262 printk(KERN_ERR
1263 "failed to allocate resource for iomem\n");
1264 return;
1267 res->name = name;
1268 res->start = md->phys_addr;
1269 res->end = md->phys_addr + efi_md_size(md) - 1;
1270 res->flags = flags;
1272 if (insert_resource(&iomem_resource, res) < 0)
1273 kfree(res);
1274 else {
1276 * We don't know which region contains
1277 * kernel data so we try it repeatedly and
1278 * let the resource manager test it.
1280 insert_resource(res, code_resource);
1281 insert_resource(res, data_resource);
1282 insert_resource(res, bss_resource);
1283 #ifdef CONFIG_KEXEC
1284 insert_resource(res, &efi_memmap_res);
1285 insert_resource(res, &boot_param_res);
1286 if (crashk_res.end > crashk_res.start)
1287 insert_resource(res, &crashk_res);
1288 #endif
1293 #ifdef CONFIG_KEXEC
1294 /* find a block of memory aligned to 64M exclude reserved regions
1295 rsvd_regions are sorted
1297 unsigned long __init
1298 kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1300 int i;
1301 u64 start, end;
1302 u64 alignment = 1UL << _PAGE_SIZE_64M;
1303 void *efi_map_start, *efi_map_end, *p;
1304 efi_memory_desc_t *md;
1305 u64 efi_desc_size;
1307 efi_map_start = __va(ia64_boot_param->efi_memmap);
1308 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1309 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1311 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1312 md = p;
1313 if (!efi_wb(md))
1314 continue;
1315 start = ALIGN(md->phys_addr, alignment);
1316 end = efi_md_end(md);
1317 for (i = 0; i < n; i++) {
1318 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1319 if (__pa(r[i].start) > start + size)
1320 return start;
1321 start = ALIGN(__pa(r[i].end), alignment);
1322 if (i < n-1 &&
1323 __pa(r[i+1].start) < start + size)
1324 continue;
1325 else
1326 break;
1329 if (end > start + size)
1330 return start;
1333 printk(KERN_WARNING
1334 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1335 return ~0UL;
1337 #endif
1339 #ifdef CONFIG_CRASH_DUMP
1340 /* locate the size find a the descriptor at a certain address */
1341 unsigned long __init
1342 vmcore_find_descriptor_size (unsigned long address)
1344 void *efi_map_start, *efi_map_end, *p;
1345 efi_memory_desc_t *md;
1346 u64 efi_desc_size;
1347 unsigned long ret = 0;
1349 efi_map_start = __va(ia64_boot_param->efi_memmap);
1350 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1351 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1353 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1354 md = p;
1355 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1356 && md->phys_addr == address) {
1357 ret = efi_md_size(md);
1358 break;
1362 if (ret == 0)
1363 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1365 return ret;
1367 #endif