xen: cleancache shim to Xen Transcendent Memory
[linux-2.6/next.git] / arch / ia64 / kernel / efi.c
blob6fc03aff046c6a133a79d1305aa7b163fd310b70
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/crash_dump.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/types.h>
30 #include <linux/slab.h>
31 #include <linux/time.h>
32 #include <linux/efi.h>
33 #include <linux/kexec.h>
34 #include <linux/mm.h>
36 #include <asm/io.h>
37 #include <asm/kregs.h>
38 #include <asm/meminit.h>
39 #include <asm/pgtable.h>
40 #include <asm/processor.h>
41 #include <asm/mca.h>
42 #include <asm/tlbflush.h>
44 #define EFI_DEBUG 0
46 extern efi_status_t efi_call_phys (void *, ...);
48 struct efi efi;
49 EXPORT_SYMBOL(efi);
50 static efi_runtime_services_t *runtime;
51 static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
53 #define efi_call_virt(f, args...) (*(f))(args)
55 #define STUB_GET_TIME(prefix, adjust_arg) \
56 static efi_status_t \
57 prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
58 { \
59 struct ia64_fpreg fr[6]; \
60 efi_time_cap_t *atc = NULL; \
61 efi_status_t ret; \
63 if (tc) \
64 atc = adjust_arg(tc); \
65 ia64_save_scratch_fpregs(fr); \
66 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \
67 adjust_arg(tm), atc); \
68 ia64_load_scratch_fpregs(fr); \
69 return ret; \
72 #define STUB_SET_TIME(prefix, adjust_arg) \
73 static efi_status_t \
74 prefix##_set_time (efi_time_t *tm) \
75 { \
76 struct ia64_fpreg fr[6]; \
77 efi_status_t ret; \
79 ia64_save_scratch_fpregs(fr); \
80 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
81 adjust_arg(tm)); \
82 ia64_load_scratch_fpregs(fr); \
83 return ret; \
86 #define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
87 static efi_status_t \
88 prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
89 efi_time_t *tm) \
90 { \
91 struct ia64_fpreg fr[6]; \
92 efi_status_t ret; \
94 ia64_save_scratch_fpregs(fr); \
95 ret = efi_call_##prefix( \
96 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
97 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
98 ia64_load_scratch_fpregs(fr); \
99 return ret; \
102 #define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
103 static efi_status_t \
104 prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
106 struct ia64_fpreg fr[6]; \
107 efi_time_t *atm = NULL; \
108 efi_status_t ret; \
110 if (tm) \
111 atm = adjust_arg(tm); \
112 ia64_save_scratch_fpregs(fr); \
113 ret = efi_call_##prefix( \
114 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
115 enabled, atm); \
116 ia64_load_scratch_fpregs(fr); \
117 return ret; \
120 #define STUB_GET_VARIABLE(prefix, adjust_arg) \
121 static efi_status_t \
122 prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
123 unsigned long *data_size, void *data) \
125 struct ia64_fpreg fr[6]; \
126 u32 *aattr = NULL; \
127 efi_status_t ret; \
129 if (attr) \
130 aattr = adjust_arg(attr); \
131 ia64_save_scratch_fpregs(fr); \
132 ret = efi_call_##prefix( \
133 (efi_get_variable_t *) __va(runtime->get_variable), \
134 adjust_arg(name), adjust_arg(vendor), aattr, \
135 adjust_arg(data_size), adjust_arg(data)); \
136 ia64_load_scratch_fpregs(fr); \
137 return ret; \
140 #define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
141 static efi_status_t \
142 prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
143 efi_guid_t *vendor) \
145 struct ia64_fpreg fr[6]; \
146 efi_status_t ret; \
148 ia64_save_scratch_fpregs(fr); \
149 ret = efi_call_##prefix( \
150 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \
151 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
152 ia64_load_scratch_fpregs(fr); \
153 return ret; \
156 #define STUB_SET_VARIABLE(prefix, adjust_arg) \
157 static efi_status_t \
158 prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
159 unsigned long attr, unsigned long data_size, \
160 void *data) \
162 struct ia64_fpreg fr[6]; \
163 efi_status_t ret; \
165 ia64_save_scratch_fpregs(fr); \
166 ret = efi_call_##prefix( \
167 (efi_set_variable_t *) __va(runtime->set_variable), \
168 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
169 adjust_arg(data)); \
170 ia64_load_scratch_fpregs(fr); \
171 return ret; \
174 #define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
175 static efi_status_t \
176 prefix##_get_next_high_mono_count (u32 *count) \
178 struct ia64_fpreg fr[6]; \
179 efi_status_t ret; \
181 ia64_save_scratch_fpregs(fr); \
182 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
183 __va(runtime->get_next_high_mono_count), \
184 adjust_arg(count)); \
185 ia64_load_scratch_fpregs(fr); \
186 return ret; \
189 #define STUB_RESET_SYSTEM(prefix, adjust_arg) \
190 static void \
191 prefix##_reset_system (int reset_type, efi_status_t status, \
192 unsigned long data_size, efi_char16_t *data) \
194 struct ia64_fpreg fr[6]; \
195 efi_char16_t *adata = NULL; \
197 if (data) \
198 adata = adjust_arg(data); \
200 ia64_save_scratch_fpregs(fr); \
201 efi_call_##prefix( \
202 (efi_reset_system_t *) __va(runtime->reset_system), \
203 reset_type, status, data_size, adata); \
204 /* should not return, but just in case... */ \
205 ia64_load_scratch_fpregs(fr); \
208 #define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
210 STUB_GET_TIME(phys, phys_ptr)
211 STUB_SET_TIME(phys, phys_ptr)
212 STUB_GET_WAKEUP_TIME(phys, phys_ptr)
213 STUB_SET_WAKEUP_TIME(phys, phys_ptr)
214 STUB_GET_VARIABLE(phys, phys_ptr)
215 STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
216 STUB_SET_VARIABLE(phys, phys_ptr)
217 STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
218 STUB_RESET_SYSTEM(phys, phys_ptr)
220 #define id(arg) arg
222 STUB_GET_TIME(virt, id)
223 STUB_SET_TIME(virt, id)
224 STUB_GET_WAKEUP_TIME(virt, id)
225 STUB_SET_WAKEUP_TIME(virt, id)
226 STUB_GET_VARIABLE(virt, id)
227 STUB_GET_NEXT_VARIABLE(virt, id)
228 STUB_SET_VARIABLE(virt, id)
229 STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
230 STUB_RESET_SYSTEM(virt, id)
232 void
233 efi_gettimeofday (struct timespec *ts)
235 efi_time_t tm;
237 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
238 memset(ts, 0, sizeof(*ts));
239 return;
242 ts->tv_sec = mktime(tm.year, tm.month, tm.day,
243 tm.hour, tm.minute, tm.second);
244 ts->tv_nsec = tm.nanosecond;
247 static int
248 is_memory_available (efi_memory_desc_t *md)
250 if (!(md->attribute & EFI_MEMORY_WB))
251 return 0;
253 switch (md->type) {
254 case EFI_LOADER_CODE:
255 case EFI_LOADER_DATA:
256 case EFI_BOOT_SERVICES_CODE:
257 case EFI_BOOT_SERVICES_DATA:
258 case EFI_CONVENTIONAL_MEMORY:
259 return 1;
261 return 0;
264 typedef struct kern_memdesc {
265 u64 attribute;
266 u64 start;
267 u64 num_pages;
268 } kern_memdesc_t;
270 static kern_memdesc_t *kern_memmap;
272 #define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
274 static inline u64
275 kmd_end(kern_memdesc_t *kmd)
277 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
280 static inline u64
281 efi_md_end(efi_memory_desc_t *md)
283 return (md->phys_addr + efi_md_size(md));
286 static inline int
287 efi_wb(efi_memory_desc_t *md)
289 return (md->attribute & EFI_MEMORY_WB);
292 static inline int
293 efi_uc(efi_memory_desc_t *md)
295 return (md->attribute & EFI_MEMORY_UC);
298 static void
299 walk (efi_freemem_callback_t callback, void *arg, u64 attr)
301 kern_memdesc_t *k;
302 u64 start, end, voff;
304 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
305 for (k = kern_memmap; k->start != ~0UL; k++) {
306 if (k->attribute != attr)
307 continue;
308 start = PAGE_ALIGN(k->start);
309 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
310 if (start < end)
311 if ((*callback)(start + voff, end + voff, arg) < 0)
312 return;
317 * Walk the EFI memory map and call CALLBACK once for each EFI memory
318 * descriptor that has memory that is available for OS use.
320 void
321 efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
323 walk(callback, arg, EFI_MEMORY_WB);
327 * Walk the EFI memory map and call CALLBACK once for each EFI memory
328 * descriptor that has memory that is available for uncached allocator.
330 void
331 efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
333 walk(callback, arg, EFI_MEMORY_UC);
337 * Look for the PAL_CODE region reported by EFI and map it using an
338 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
339 * Abstraction Layer chapter 11 in ADAG
341 void *
342 efi_get_pal_addr (void)
344 void *efi_map_start, *efi_map_end, *p;
345 efi_memory_desc_t *md;
346 u64 efi_desc_size;
347 int pal_code_count = 0;
348 u64 vaddr, mask;
350 efi_map_start = __va(ia64_boot_param->efi_memmap);
351 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
352 efi_desc_size = ia64_boot_param->efi_memdesc_size;
354 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
355 md = p;
356 if (md->type != EFI_PAL_CODE)
357 continue;
359 if (++pal_code_count > 1) {
360 printk(KERN_ERR "Too many EFI Pal Code memory ranges, "
361 "dropped @ %llx\n", md->phys_addr);
362 continue;
365 * The only ITLB entry in region 7 that is used is the one
366 * installed by __start(). That entry covers a 64MB range.
368 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
369 vaddr = PAGE_OFFSET + md->phys_addr;
372 * We must check that the PAL mapping won't overlap with the
373 * kernel mapping.
375 * PAL code is guaranteed to be aligned on a power of 2 between
376 * 4k and 256KB and that only one ITR is needed to map it. This
377 * implies that the PAL code is always aligned on its size,
378 * i.e., the closest matching page size supported by the TLB.
379 * Therefore PAL code is guaranteed never to cross a 64MB unless
380 * it is bigger than 64MB (very unlikely!). So for now the
381 * following test is enough to determine whether or not we need
382 * a dedicated ITR for the PAL code.
384 if ((vaddr & mask) == (KERNEL_START & mask)) {
385 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
386 __func__);
387 continue;
390 if (efi_md_size(md) > IA64_GRANULE_SIZE)
391 panic("Whoa! PAL code size bigger than a granule!");
393 #if EFI_DEBUG
394 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
396 printk(KERN_INFO "CPU %d: mapping PAL code "
397 "[0x%lx-0x%lx) into [0x%lx-0x%lx)\n",
398 smp_processor_id(), md->phys_addr,
399 md->phys_addr + efi_md_size(md),
400 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
401 #endif
402 return __va(md->phys_addr);
404 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
405 __func__);
406 return NULL;
410 static u8 __init palo_checksum(u8 *buffer, u32 length)
412 u8 sum = 0;
413 u8 *end = buffer + length;
415 while (buffer < end)
416 sum = (u8) (sum + *(buffer++));
418 return sum;
422 * Parse and handle PALO table which is published at:
423 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
425 static void __init handle_palo(unsigned long palo_phys)
427 struct palo_table *palo = __va(palo_phys);
428 u8 checksum;
430 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
431 printk(KERN_INFO "PALO signature incorrect.\n");
432 return;
435 checksum = palo_checksum((u8 *)palo, palo->length);
436 if (checksum) {
437 printk(KERN_INFO "PALO checksum incorrect.\n");
438 return;
441 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
444 void
445 efi_map_pal_code (void)
447 void *pal_vaddr = efi_get_pal_addr ();
448 u64 psr;
450 if (!pal_vaddr)
451 return;
454 * Cannot write to CRx with PSR.ic=1
456 psr = ia64_clear_ic();
457 ia64_itr(0x1, IA64_TR_PALCODE,
458 GRANULEROUNDDOWN((unsigned long) pal_vaddr),
459 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
460 IA64_GRANULE_SHIFT);
461 paravirt_dv_serialize_data();
462 ia64_set_psr(psr); /* restore psr */
465 void __init
466 efi_init (void)
468 void *efi_map_start, *efi_map_end;
469 efi_config_table_t *config_tables;
470 efi_char16_t *c16;
471 u64 efi_desc_size;
472 char *cp, vendor[100] = "unknown";
473 int i;
474 unsigned long palo_phys;
477 * It's too early to be able to use the standard kernel command line
478 * support...
480 for (cp = boot_command_line; *cp; ) {
481 if (memcmp(cp, "mem=", 4) == 0) {
482 mem_limit = memparse(cp + 4, &cp);
483 } else if (memcmp(cp, "max_addr=", 9) == 0) {
484 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
485 } else if (memcmp(cp, "min_addr=", 9) == 0) {
486 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
487 } else {
488 while (*cp != ' ' && *cp)
489 ++cp;
490 while (*cp == ' ')
491 ++cp;
494 if (min_addr != 0UL)
495 printk(KERN_INFO "Ignoring memory below %lluMB\n",
496 min_addr >> 20);
497 if (max_addr != ~0UL)
498 printk(KERN_INFO "Ignoring memory above %lluMB\n",
499 max_addr >> 20);
501 efi.systab = __va(ia64_boot_param->efi_systab);
504 * Verify the EFI Table
506 if (efi.systab == NULL)
507 panic("Whoa! Can't find EFI system table.\n");
508 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
509 panic("Whoa! EFI system table signature incorrect\n");
510 if ((efi.systab->hdr.revision >> 16) == 0)
511 printk(KERN_WARNING "Warning: EFI system table version "
512 "%d.%02d, expected 1.00 or greater\n",
513 efi.systab->hdr.revision >> 16,
514 efi.systab->hdr.revision & 0xffff);
516 config_tables = __va(efi.systab->tables);
518 /* Show what we know for posterity */
519 c16 = __va(efi.systab->fw_vendor);
520 if (c16) {
521 for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i)
522 vendor[i] = *c16++;
523 vendor[i] = '\0';
526 printk(KERN_INFO "EFI v%u.%.02u by %s:",
527 efi.systab->hdr.revision >> 16,
528 efi.systab->hdr.revision & 0xffff, vendor);
530 efi.mps = EFI_INVALID_TABLE_ADDR;
531 efi.acpi = EFI_INVALID_TABLE_ADDR;
532 efi.acpi20 = EFI_INVALID_TABLE_ADDR;
533 efi.smbios = EFI_INVALID_TABLE_ADDR;
534 efi.sal_systab = EFI_INVALID_TABLE_ADDR;
535 efi.boot_info = EFI_INVALID_TABLE_ADDR;
536 efi.hcdp = EFI_INVALID_TABLE_ADDR;
537 efi.uga = EFI_INVALID_TABLE_ADDR;
539 palo_phys = EFI_INVALID_TABLE_ADDR;
541 for (i = 0; i < (int) efi.systab->nr_tables; i++) {
542 if (efi_guidcmp(config_tables[i].guid, MPS_TABLE_GUID) == 0) {
543 efi.mps = config_tables[i].table;
544 printk(" MPS=0x%lx", config_tables[i].table);
545 } else if (efi_guidcmp(config_tables[i].guid, ACPI_20_TABLE_GUID) == 0) {
546 efi.acpi20 = config_tables[i].table;
547 printk(" ACPI 2.0=0x%lx", config_tables[i].table);
548 } else if (efi_guidcmp(config_tables[i].guid, ACPI_TABLE_GUID) == 0) {
549 efi.acpi = config_tables[i].table;
550 printk(" ACPI=0x%lx", config_tables[i].table);
551 } else if (efi_guidcmp(config_tables[i].guid, SMBIOS_TABLE_GUID) == 0) {
552 efi.smbios = config_tables[i].table;
553 printk(" SMBIOS=0x%lx", config_tables[i].table);
554 } else if (efi_guidcmp(config_tables[i].guid, SAL_SYSTEM_TABLE_GUID) == 0) {
555 efi.sal_systab = config_tables[i].table;
556 printk(" SALsystab=0x%lx", config_tables[i].table);
557 } else if (efi_guidcmp(config_tables[i].guid, HCDP_TABLE_GUID) == 0) {
558 efi.hcdp = config_tables[i].table;
559 printk(" HCDP=0x%lx", config_tables[i].table);
560 } else if (efi_guidcmp(config_tables[i].guid,
561 PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID) == 0) {
562 palo_phys = config_tables[i].table;
563 printk(" PALO=0x%lx", config_tables[i].table);
566 printk("\n");
568 if (palo_phys != EFI_INVALID_TABLE_ADDR)
569 handle_palo(palo_phys);
571 runtime = __va(efi.systab->runtime);
572 efi.get_time = phys_get_time;
573 efi.set_time = phys_set_time;
574 efi.get_wakeup_time = phys_get_wakeup_time;
575 efi.set_wakeup_time = phys_set_wakeup_time;
576 efi.get_variable = phys_get_variable;
577 efi.get_next_variable = phys_get_next_variable;
578 efi.set_variable = phys_set_variable;
579 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
580 efi.reset_system = phys_reset_system;
582 efi_map_start = __va(ia64_boot_param->efi_memmap);
583 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
584 efi_desc_size = ia64_boot_param->efi_memdesc_size;
586 #if EFI_DEBUG
587 /* print EFI memory map: */
589 efi_memory_desc_t *md;
590 void *p;
592 for (i = 0, p = efi_map_start; p < efi_map_end;
593 ++i, p += efi_desc_size)
595 const char *unit;
596 unsigned long size;
598 md = p;
599 size = md->num_pages << EFI_PAGE_SHIFT;
601 if ((size >> 40) > 0) {
602 size >>= 40;
603 unit = "TB";
604 } else if ((size >> 30) > 0) {
605 size >>= 30;
606 unit = "GB";
607 } else if ((size >> 20) > 0) {
608 size >>= 20;
609 unit = "MB";
610 } else {
611 size >>= 10;
612 unit = "KB";
615 printk("mem%02d: type=%2u, attr=0x%016lx, "
616 "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
617 i, md->type, md->attribute, md->phys_addr,
618 md->phys_addr + efi_md_size(md), size, unit);
621 #endif
623 efi_map_pal_code();
624 efi_enter_virtual_mode();
627 void
628 efi_enter_virtual_mode (void)
630 void *efi_map_start, *efi_map_end, *p;
631 efi_memory_desc_t *md;
632 efi_status_t status;
633 u64 efi_desc_size;
635 efi_map_start = __va(ia64_boot_param->efi_memmap);
636 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
637 efi_desc_size = ia64_boot_param->efi_memdesc_size;
639 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
640 md = p;
641 if (md->attribute & EFI_MEMORY_RUNTIME) {
643 * Some descriptors have multiple bits set, so the
644 * order of the tests is relevant.
646 if (md->attribute & EFI_MEMORY_WB) {
647 md->virt_addr = (u64) __va(md->phys_addr);
648 } else if (md->attribute & EFI_MEMORY_UC) {
649 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
650 } else if (md->attribute & EFI_MEMORY_WC) {
651 #if 0
652 md->virt_addr = ia64_remap(md->phys_addr,
653 (_PAGE_A |
654 _PAGE_P |
655 _PAGE_D |
656 _PAGE_MA_WC |
657 _PAGE_PL_0 |
658 _PAGE_AR_RW));
659 #else
660 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
661 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
662 #endif
663 } else if (md->attribute & EFI_MEMORY_WT) {
664 #if 0
665 md->virt_addr = ia64_remap(md->phys_addr,
666 (_PAGE_A |
667 _PAGE_P |
668 _PAGE_D |
669 _PAGE_MA_WT |
670 _PAGE_PL_0 |
671 _PAGE_AR_RW));
672 #else
673 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
674 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
675 #endif
680 status = efi_call_phys(__va(runtime->set_virtual_address_map),
681 ia64_boot_param->efi_memmap_size,
682 efi_desc_size,
683 ia64_boot_param->efi_memdesc_version,
684 ia64_boot_param->efi_memmap);
685 if (status != EFI_SUCCESS) {
686 printk(KERN_WARNING "warning: unable to switch EFI into "
687 "virtual mode (status=%lu)\n", status);
688 return;
692 * Now that EFI is in virtual mode, we call the EFI functions more
693 * efficiently:
695 efi.get_time = virt_get_time;
696 efi.set_time = virt_set_time;
697 efi.get_wakeup_time = virt_get_wakeup_time;
698 efi.set_wakeup_time = virt_set_wakeup_time;
699 efi.get_variable = virt_get_variable;
700 efi.get_next_variable = virt_get_next_variable;
701 efi.set_variable = virt_set_variable;
702 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
703 efi.reset_system = virt_reset_system;
707 * Walk the EFI memory map looking for the I/O port range. There can only be
708 * one entry of this type, other I/O port ranges should be described via ACPI.
711 efi_get_iobase (void)
713 void *efi_map_start, *efi_map_end, *p;
714 efi_memory_desc_t *md;
715 u64 efi_desc_size;
717 efi_map_start = __va(ia64_boot_param->efi_memmap);
718 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
719 efi_desc_size = ia64_boot_param->efi_memdesc_size;
721 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
722 md = p;
723 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
724 if (md->attribute & EFI_MEMORY_UC)
725 return md->phys_addr;
728 return 0;
731 static struct kern_memdesc *
732 kern_memory_descriptor (unsigned long phys_addr)
734 struct kern_memdesc *md;
736 for (md = kern_memmap; md->start != ~0UL; md++) {
737 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
738 return md;
740 return NULL;
743 static efi_memory_desc_t *
744 efi_memory_descriptor (unsigned long phys_addr)
746 void *efi_map_start, *efi_map_end, *p;
747 efi_memory_desc_t *md;
748 u64 efi_desc_size;
750 efi_map_start = __va(ia64_boot_param->efi_memmap);
751 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
752 efi_desc_size = ia64_boot_param->efi_memdesc_size;
754 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
755 md = p;
757 if (phys_addr - md->phys_addr < efi_md_size(md))
758 return md;
760 return NULL;
763 static int
764 efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
766 void *efi_map_start, *efi_map_end, *p;
767 efi_memory_desc_t *md;
768 u64 efi_desc_size;
769 unsigned long end;
771 efi_map_start = __va(ia64_boot_param->efi_memmap);
772 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
773 efi_desc_size = ia64_boot_param->efi_memdesc_size;
775 end = phys_addr + size;
777 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
778 md = p;
779 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
780 return 1;
782 return 0;
786 efi_mem_type (unsigned long phys_addr)
788 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
790 if (md)
791 return md->type;
792 return 0;
796 efi_mem_attributes (unsigned long phys_addr)
798 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
800 if (md)
801 return md->attribute;
802 return 0;
804 EXPORT_SYMBOL(efi_mem_attributes);
807 efi_mem_attribute (unsigned long phys_addr, unsigned long size)
809 unsigned long end = phys_addr + size;
810 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
811 u64 attr;
813 if (!md)
814 return 0;
817 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
818 * the kernel that firmware needs this region mapped.
820 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
821 do {
822 unsigned long md_end = efi_md_end(md);
824 if (end <= md_end)
825 return attr;
827 md = efi_memory_descriptor(md_end);
828 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
829 return 0;
830 } while (md);
831 return 0; /* never reached */
835 kern_mem_attribute (unsigned long phys_addr, unsigned long size)
837 unsigned long end = phys_addr + size;
838 struct kern_memdesc *md;
839 u64 attr;
842 * This is a hack for ioremap calls before we set up kern_memmap.
843 * Maybe we should do efi_memmap_init() earlier instead.
845 if (!kern_memmap) {
846 attr = efi_mem_attribute(phys_addr, size);
847 if (attr & EFI_MEMORY_WB)
848 return EFI_MEMORY_WB;
849 return 0;
852 md = kern_memory_descriptor(phys_addr);
853 if (!md)
854 return 0;
856 attr = md->attribute;
857 do {
858 unsigned long md_end = kmd_end(md);
860 if (end <= md_end)
861 return attr;
863 md = kern_memory_descriptor(md_end);
864 if (!md || md->attribute != attr)
865 return 0;
866 } while (md);
867 return 0; /* never reached */
869 EXPORT_SYMBOL(kern_mem_attribute);
872 valid_phys_addr_range (unsigned long phys_addr, unsigned long size)
874 u64 attr;
877 * /dev/mem reads and writes use copy_to_user(), which implicitly
878 * uses a granule-sized kernel identity mapping. It's really
879 * only safe to do this for regions in kern_memmap. For more
880 * details, see Documentation/ia64/aliasing.txt.
882 attr = kern_mem_attribute(phys_addr, size);
883 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
884 return 1;
885 return 0;
889 valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
891 unsigned long phys_addr = pfn << PAGE_SHIFT;
892 u64 attr;
894 attr = efi_mem_attribute(phys_addr, size);
897 * /dev/mem mmap uses normal user pages, so we don't need the entire
898 * granule, but the entire region we're mapping must support the same
899 * attribute.
901 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
902 return 1;
905 * Intel firmware doesn't tell us about all the MMIO regions, so
906 * in general we have to allow mmap requests. But if EFI *does*
907 * tell us about anything inside this region, we should deny it.
908 * The user can always map a smaller region to avoid the overlap.
910 if (efi_memmap_intersects(phys_addr, size))
911 return 0;
913 return 1;
916 pgprot_t
917 phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
918 pgprot_t vma_prot)
920 unsigned long phys_addr = pfn << PAGE_SHIFT;
921 u64 attr;
924 * For /dev/mem mmap, we use user mappings, but if the region is
925 * in kern_memmap (and hence may be covered by a kernel mapping),
926 * we must use the same attribute as the kernel mapping.
928 attr = kern_mem_attribute(phys_addr, size);
929 if (attr & EFI_MEMORY_WB)
930 return pgprot_cacheable(vma_prot);
931 else if (attr & EFI_MEMORY_UC)
932 return pgprot_noncached(vma_prot);
935 * Some chipsets don't support UC access to memory. If
936 * WB is supported, we prefer that.
938 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
939 return pgprot_cacheable(vma_prot);
941 return pgprot_noncached(vma_prot);
944 int __init
945 efi_uart_console_only(void)
947 efi_status_t status;
948 char *s, name[] = "ConOut";
949 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
950 efi_char16_t *utf16, name_utf16[32];
951 unsigned char data[1024];
952 unsigned long size = sizeof(data);
953 struct efi_generic_dev_path *hdr, *end_addr;
954 int uart = 0;
956 /* Convert to UTF-16 */
957 utf16 = name_utf16;
958 s = name;
959 while (*s)
960 *utf16++ = *s++ & 0x7f;
961 *utf16 = 0;
963 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
964 if (status != EFI_SUCCESS) {
965 printk(KERN_ERR "No EFI %s variable?\n", name);
966 return 0;
969 hdr = (struct efi_generic_dev_path *) data;
970 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
971 while (hdr < end_addr) {
972 if (hdr->type == EFI_DEV_MSG &&
973 hdr->sub_type == EFI_DEV_MSG_UART)
974 uart = 1;
975 else if (hdr->type == EFI_DEV_END_PATH ||
976 hdr->type == EFI_DEV_END_PATH2) {
977 if (!uart)
978 return 0;
979 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
980 return 1;
981 uart = 0;
983 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
985 printk(KERN_ERR "Malformed %s value\n", name);
986 return 0;
990 * Look for the first granule aligned memory descriptor memory
991 * that is big enough to hold EFI memory map. Make sure this
992 * descriptor is atleast granule sized so it does not get trimmed
994 struct kern_memdesc *
995 find_memmap_space (void)
997 u64 contig_low=0, contig_high=0;
998 u64 as = 0, ae;
999 void *efi_map_start, *efi_map_end, *p, *q;
1000 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1001 u64 space_needed, efi_desc_size;
1002 unsigned long total_mem = 0;
1004 efi_map_start = __va(ia64_boot_param->efi_memmap);
1005 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1006 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1009 * Worst case: we need 3 kernel descriptors for each efi descriptor
1010 * (if every entry has a WB part in the middle, and UC head and tail),
1011 * plus one for the end marker.
1013 space_needed = sizeof(kern_memdesc_t) *
1014 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
1016 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1017 md = p;
1018 if (!efi_wb(md)) {
1019 continue;
1021 if (pmd == NULL || !efi_wb(pmd) ||
1022 efi_md_end(pmd) != md->phys_addr) {
1023 contig_low = GRANULEROUNDUP(md->phys_addr);
1024 contig_high = efi_md_end(md);
1025 for (q = p + efi_desc_size; q < efi_map_end;
1026 q += efi_desc_size) {
1027 check_md = q;
1028 if (!efi_wb(check_md))
1029 break;
1030 if (contig_high != check_md->phys_addr)
1031 break;
1032 contig_high = efi_md_end(check_md);
1034 contig_high = GRANULEROUNDDOWN(contig_high);
1036 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
1037 continue;
1039 /* Round ends inward to granule boundaries */
1040 as = max(contig_low, md->phys_addr);
1041 ae = min(contig_high, efi_md_end(md));
1043 /* keep within max_addr= and min_addr= command line arg */
1044 as = max(as, min_addr);
1045 ae = min(ae, max_addr);
1046 if (ae <= as)
1047 continue;
1049 /* avoid going over mem= command line arg */
1050 if (total_mem + (ae - as) > mem_limit)
1051 ae -= total_mem + (ae - as) - mem_limit;
1053 if (ae <= as)
1054 continue;
1056 if (ae - as > space_needed)
1057 break;
1059 if (p >= efi_map_end)
1060 panic("Can't allocate space for kernel memory descriptors");
1062 return __va(as);
1066 * Walk the EFI memory map and gather all memory available for kernel
1067 * to use. We can allocate partial granules only if the unavailable
1068 * parts exist, and are WB.
1070 unsigned long
1071 efi_memmap_init(u64 *s, u64 *e)
1073 struct kern_memdesc *k, *prev = NULL;
1074 u64 contig_low=0, contig_high=0;
1075 u64 as, ae, lim;
1076 void *efi_map_start, *efi_map_end, *p, *q;
1077 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1078 u64 efi_desc_size;
1079 unsigned long total_mem = 0;
1081 k = kern_memmap = find_memmap_space();
1083 efi_map_start = __va(ia64_boot_param->efi_memmap);
1084 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1085 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1087 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1088 md = p;
1089 if (!efi_wb(md)) {
1090 if (efi_uc(md) &&
1091 (md->type == EFI_CONVENTIONAL_MEMORY ||
1092 md->type == EFI_BOOT_SERVICES_DATA)) {
1093 k->attribute = EFI_MEMORY_UC;
1094 k->start = md->phys_addr;
1095 k->num_pages = md->num_pages;
1096 k++;
1098 continue;
1100 if (pmd == NULL || !efi_wb(pmd) ||
1101 efi_md_end(pmd) != md->phys_addr) {
1102 contig_low = GRANULEROUNDUP(md->phys_addr);
1103 contig_high = efi_md_end(md);
1104 for (q = p + efi_desc_size; q < efi_map_end;
1105 q += efi_desc_size) {
1106 check_md = q;
1107 if (!efi_wb(check_md))
1108 break;
1109 if (contig_high != check_md->phys_addr)
1110 break;
1111 contig_high = efi_md_end(check_md);
1113 contig_high = GRANULEROUNDDOWN(contig_high);
1115 if (!is_memory_available(md))
1116 continue;
1118 #ifdef CONFIG_CRASH_DUMP
1119 /* saved_max_pfn should ignore max_addr= command line arg */
1120 if (saved_max_pfn < (efi_md_end(md) >> PAGE_SHIFT))
1121 saved_max_pfn = (efi_md_end(md) >> PAGE_SHIFT);
1122 #endif
1124 * Round ends inward to granule boundaries
1125 * Give trimmings to uncached allocator
1127 if (md->phys_addr < contig_low) {
1128 lim = min(efi_md_end(md), contig_low);
1129 if (efi_uc(md)) {
1130 if (k > kern_memmap &&
1131 (k-1)->attribute == EFI_MEMORY_UC &&
1132 kmd_end(k-1) == md->phys_addr) {
1133 (k-1)->num_pages +=
1134 (lim - md->phys_addr)
1135 >> EFI_PAGE_SHIFT;
1136 } else {
1137 k->attribute = EFI_MEMORY_UC;
1138 k->start = md->phys_addr;
1139 k->num_pages = (lim - md->phys_addr)
1140 >> EFI_PAGE_SHIFT;
1141 k++;
1144 as = contig_low;
1145 } else
1146 as = md->phys_addr;
1148 if (efi_md_end(md) > contig_high) {
1149 lim = max(md->phys_addr, contig_high);
1150 if (efi_uc(md)) {
1151 if (lim == md->phys_addr && k > kern_memmap &&
1152 (k-1)->attribute == EFI_MEMORY_UC &&
1153 kmd_end(k-1) == md->phys_addr) {
1154 (k-1)->num_pages += md->num_pages;
1155 } else {
1156 k->attribute = EFI_MEMORY_UC;
1157 k->start = lim;
1158 k->num_pages = (efi_md_end(md) - lim)
1159 >> EFI_PAGE_SHIFT;
1160 k++;
1163 ae = contig_high;
1164 } else
1165 ae = efi_md_end(md);
1167 /* keep within max_addr= and min_addr= command line arg */
1168 as = max(as, min_addr);
1169 ae = min(ae, max_addr);
1170 if (ae <= as)
1171 continue;
1173 /* avoid going over mem= command line arg */
1174 if (total_mem + (ae - as) > mem_limit)
1175 ae -= total_mem + (ae - as) - mem_limit;
1177 if (ae <= as)
1178 continue;
1179 if (prev && kmd_end(prev) == md->phys_addr) {
1180 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1181 total_mem += ae - as;
1182 continue;
1184 k->attribute = EFI_MEMORY_WB;
1185 k->start = as;
1186 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1187 total_mem += ae - as;
1188 prev = k++;
1190 k->start = ~0L; /* end-marker */
1192 /* reserve the memory we are using for kern_memmap */
1193 *s = (u64)kern_memmap;
1194 *e = (u64)++k;
1196 return total_mem;
1199 void
1200 efi_initialize_iomem_resources(struct resource *code_resource,
1201 struct resource *data_resource,
1202 struct resource *bss_resource)
1204 struct resource *res;
1205 void *efi_map_start, *efi_map_end, *p;
1206 efi_memory_desc_t *md;
1207 u64 efi_desc_size;
1208 char *name;
1209 unsigned long flags;
1211 efi_map_start = __va(ia64_boot_param->efi_memmap);
1212 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1213 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1215 res = NULL;
1217 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1218 md = p;
1220 if (md->num_pages == 0) /* should not happen */
1221 continue;
1223 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1224 switch (md->type) {
1226 case EFI_MEMORY_MAPPED_IO:
1227 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1228 continue;
1230 case EFI_LOADER_CODE:
1231 case EFI_LOADER_DATA:
1232 case EFI_BOOT_SERVICES_DATA:
1233 case EFI_BOOT_SERVICES_CODE:
1234 case EFI_CONVENTIONAL_MEMORY:
1235 if (md->attribute & EFI_MEMORY_WP) {
1236 name = "System ROM";
1237 flags |= IORESOURCE_READONLY;
1238 } else if (md->attribute == EFI_MEMORY_UC)
1239 name = "Uncached RAM";
1240 else
1241 name = "System RAM";
1242 break;
1244 case EFI_ACPI_MEMORY_NVS:
1245 name = "ACPI Non-volatile Storage";
1246 break;
1248 case EFI_UNUSABLE_MEMORY:
1249 name = "reserved";
1250 flags |= IORESOURCE_DISABLED;
1251 break;
1253 case EFI_RESERVED_TYPE:
1254 case EFI_RUNTIME_SERVICES_CODE:
1255 case EFI_RUNTIME_SERVICES_DATA:
1256 case EFI_ACPI_RECLAIM_MEMORY:
1257 default:
1258 name = "reserved";
1259 break;
1262 if ((res = kzalloc(sizeof(struct resource),
1263 GFP_KERNEL)) == NULL) {
1264 printk(KERN_ERR
1265 "failed to allocate resource for iomem\n");
1266 return;
1269 res->name = name;
1270 res->start = md->phys_addr;
1271 res->end = md->phys_addr + efi_md_size(md) - 1;
1272 res->flags = flags;
1274 if (insert_resource(&iomem_resource, res) < 0)
1275 kfree(res);
1276 else {
1278 * We don't know which region contains
1279 * kernel data so we try it repeatedly and
1280 * let the resource manager test it.
1282 insert_resource(res, code_resource);
1283 insert_resource(res, data_resource);
1284 insert_resource(res, bss_resource);
1285 #ifdef CONFIG_KEXEC
1286 insert_resource(res, &efi_memmap_res);
1287 insert_resource(res, &boot_param_res);
1288 if (crashk_res.end > crashk_res.start)
1289 insert_resource(res, &crashk_res);
1290 #endif
1295 #ifdef CONFIG_KEXEC
1296 /* find a block of memory aligned to 64M exclude reserved regions
1297 rsvd_regions are sorted
1299 unsigned long __init
1300 kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1302 int i;
1303 u64 start, end;
1304 u64 alignment = 1UL << _PAGE_SIZE_64M;
1305 void *efi_map_start, *efi_map_end, *p;
1306 efi_memory_desc_t *md;
1307 u64 efi_desc_size;
1309 efi_map_start = __va(ia64_boot_param->efi_memmap);
1310 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1311 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1313 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1314 md = p;
1315 if (!efi_wb(md))
1316 continue;
1317 start = ALIGN(md->phys_addr, alignment);
1318 end = efi_md_end(md);
1319 for (i = 0; i < n; i++) {
1320 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1321 if (__pa(r[i].start) > start + size)
1322 return start;
1323 start = ALIGN(__pa(r[i].end), alignment);
1324 if (i < n-1 &&
1325 __pa(r[i+1].start) < start + size)
1326 continue;
1327 else
1328 break;
1331 if (end > start + size)
1332 return start;
1335 printk(KERN_WARNING
1336 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1337 return ~0UL;
1339 #endif
1341 #ifdef CONFIG_CRASH_DUMP
1342 /* locate the size find a the descriptor at a certain address */
1343 unsigned long __init
1344 vmcore_find_descriptor_size (unsigned long address)
1346 void *efi_map_start, *efi_map_end, *p;
1347 efi_memory_desc_t *md;
1348 u64 efi_desc_size;
1349 unsigned long ret = 0;
1351 efi_map_start = __va(ia64_boot_param->efi_memmap);
1352 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1353 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1355 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1356 md = p;
1357 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1358 && md->phys_addr == address) {
1359 ret = efi_md_size(md);
1360 break;
1364 if (ret == 0)
1365 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1367 return ret;
1369 #endif