2 * Extensible Firmware Interface
4 * Based on Extensible Firmware Interface Specification version 0.9
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>
35 #include <asm/kregs.h>
36 #include <asm/meminit.h>
37 #include <asm/pgtable.h>
38 #include <asm/processor.h>
40 #include <asm/tlbflush.h>
44 extern efi_status_t
efi_call_phys (void *, ...);
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) \
55 prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
57 struct ia64_fpreg fr[6]; \
58 efi_time_cap_t *atc = NULL; \
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); \
70 #define STUB_SET_TIME(prefix, adjust_arg) \
72 prefix##_set_time (efi_time_t *tm) \
74 struct ia64_fpreg fr[6]; \
77 ia64_save_scratch_fpregs(fr); \
78 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
80 ia64_load_scratch_fpregs(fr); \
84 #define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
86 prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
89 struct ia64_fpreg fr[6]; \
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); \
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; \
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), \
114 ia64_load_scratch_fpregs(fr); \
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]; \
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); \
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]; \
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); \
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, \
160 struct ia64_fpreg fr[6]; \
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, \
168 ia64_load_scratch_fpregs(fr); \
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]; \
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); \
187 #define STUB_RESET_SYSTEM(prefix, adjust_arg) \
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; \
196 adata = adjust_arg(data); \
198 ia64_save_scratch_fpregs(fr); \
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
)
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
)
231 efi_gettimeofday (struct timespec
*ts
)
235 if ((*efi
.get_time
)(&tm
, NULL
) != EFI_SUCCESS
) {
236 memset(ts
, 0, sizeof(*ts
));
240 ts
->tv_sec
= mktime(tm
.year
, tm
.month
, tm
.day
,
241 tm
.hour
, tm
.minute
, tm
.second
);
242 ts
->tv_nsec
= tm
.nanosecond
;
246 is_memory_available (efi_memory_desc_t
*md
)
248 if (!(md
->attribute
& EFI_MEMORY_WB
))
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
:
262 typedef struct kern_memdesc
{
268 static kern_memdesc_t
*kern_memmap
;
270 #define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
273 kmd_end(kern_memdesc_t
*kmd
)
275 return (kmd
->start
+ (kmd
->num_pages
<< EFI_PAGE_SHIFT
));
279 efi_md_end(efi_memory_desc_t
*md
)
281 return (md
->phys_addr
+ efi_md_size(md
));
285 efi_wb(efi_memory_desc_t
*md
)
287 return (md
->attribute
& EFI_MEMORY_WB
);
291 efi_uc(efi_memory_desc_t
*md
)
293 return (md
->attribute
& EFI_MEMORY_UC
);
297 walk (efi_freemem_callback_t callback
, void *arg
, u64 attr
)
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
)
306 start
= PAGE_ALIGN(k
->start
);
307 end
= (k
->start
+ (k
->num_pages
<< EFI_PAGE_SHIFT
)) & PAGE_MASK
;
309 if ((*callback
)(start
+ voff
, end
+ voff
, arg
) < 0)
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.
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.
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
340 efi_get_pal_addr (void)
342 void *efi_map_start
, *efi_map_end
, *p
;
343 efi_memory_desc_t
*md
;
345 int pal_code_count
= 0;
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
) {
354 if (md
->type
!= EFI_PAL_CODE
)
357 if (++pal_code_count
> 1) {
358 printk(KERN_ERR
"Too many EFI Pal Code memory ranges, "
359 "dropped @ %lx\n", md
->phys_addr
);
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
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",
388 if (efi_md_size(md
) > IA64_GRANULE_SIZE
)
389 panic("Whoa! PAL code size bigger than a granule!");
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
);
400 return __va(md
->phys_addr
);
402 printk(KERN_WARNING
"%s: no PAL-code memory-descriptor found\n",
408 static u8 __init
palo_checksum(u8
*buffer
, u32 length
)
411 u8
*end
= buffer
+ length
;
414 sum
= (u8
) (sum
+ *(buffer
++));
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
);
428 if (strncmp(palo
->signature
, PALO_SIG
, sizeof(PALO_SIG
) - 1)) {
429 printk(KERN_INFO
"PALO signature incorrect.\n");
433 checksum
= palo_checksum((u8
*)palo
, palo
->length
);
435 printk(KERN_INFO
"PALO checksum incorrect.\n");
439 setup_ptcg_sem(palo
->max_tlb_purges
, NPTCG_FROM_PALO
);
443 efi_map_pal_code (void)
445 void *pal_vaddr
= efi_get_pal_addr ();
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
)),
459 ia64_set_psr(psr
); /* restore psr */
465 void *efi_map_start
, *efi_map_end
;
466 efi_config_table_t
*config_tables
;
469 char *cp
, vendor
[100] = "unknown";
471 unsigned long palo_phys
;
474 * It's too early to be able to use the standard kernel command line
477 for (cp
= boot_command_line
; *cp
; ) {
478 if (memcmp(cp
, "mem=", 4) == 0) {
479 mem_limit
= memparse(cp
+ 4, &cp
);
480 } else if (memcmp(cp
, "max_addr=", 9) == 0) {
481 max_addr
= GRANULEROUNDDOWN(memparse(cp
+ 9, &cp
));
482 } else if (memcmp(cp
, "min_addr=", 9) == 0) {
483 min_addr
= GRANULEROUNDDOWN(memparse(cp
+ 9, &cp
));
485 while (*cp
!= ' ' && *cp
)
492 printk(KERN_INFO
"Ignoring memory below %luMB\n",
494 if (max_addr
!= ~0UL)
495 printk(KERN_INFO
"Ignoring memory above %luMB\n",
498 efi
.systab
= __va(ia64_boot_param
->efi_systab
);
501 * Verify the EFI Table
503 if (efi
.systab
== NULL
)
504 panic("Whoa! Can't find EFI system table.\n");
505 if (efi
.systab
->hdr
.signature
!= EFI_SYSTEM_TABLE_SIGNATURE
)
506 panic("Whoa! EFI system table signature incorrect\n");
507 if ((efi
.systab
->hdr
.revision
>> 16) == 0)
508 printk(KERN_WARNING
"Warning: EFI system table version "
509 "%d.%02d, expected 1.00 or greater\n",
510 efi
.systab
->hdr
.revision
>> 16,
511 efi
.systab
->hdr
.revision
& 0xffff);
513 config_tables
= __va(efi
.systab
->tables
);
515 /* Show what we know for posterity */
516 c16
= __va(efi
.systab
->fw_vendor
);
518 for (i
= 0;i
< (int) sizeof(vendor
) - 1 && *c16
; ++i
)
523 printk(KERN_INFO
"EFI v%u.%.02u by %s:",
524 efi
.systab
->hdr
.revision
>> 16,
525 efi
.systab
->hdr
.revision
& 0xffff, vendor
);
527 efi
.mps
= EFI_INVALID_TABLE_ADDR
;
528 efi
.acpi
= EFI_INVALID_TABLE_ADDR
;
529 efi
.acpi20
= EFI_INVALID_TABLE_ADDR
;
530 efi
.smbios
= EFI_INVALID_TABLE_ADDR
;
531 efi
.sal_systab
= EFI_INVALID_TABLE_ADDR
;
532 efi
.boot_info
= EFI_INVALID_TABLE_ADDR
;
533 efi
.hcdp
= EFI_INVALID_TABLE_ADDR
;
534 efi
.uga
= EFI_INVALID_TABLE_ADDR
;
536 palo_phys
= EFI_INVALID_TABLE_ADDR
;
538 for (i
= 0; i
< (int) efi
.systab
->nr_tables
; i
++) {
539 if (efi_guidcmp(config_tables
[i
].guid
, MPS_TABLE_GUID
) == 0) {
540 efi
.mps
= config_tables
[i
].table
;
541 printk(" MPS=0x%lx", config_tables
[i
].table
);
542 } else if (efi_guidcmp(config_tables
[i
].guid
, ACPI_20_TABLE_GUID
) == 0) {
543 efi
.acpi20
= config_tables
[i
].table
;
544 printk(" ACPI 2.0=0x%lx", config_tables
[i
].table
);
545 } else if (efi_guidcmp(config_tables
[i
].guid
, ACPI_TABLE_GUID
) == 0) {
546 efi
.acpi
= config_tables
[i
].table
;
547 printk(" ACPI=0x%lx", config_tables
[i
].table
);
548 } else if (efi_guidcmp(config_tables
[i
].guid
, SMBIOS_TABLE_GUID
) == 0) {
549 efi
.smbios
= config_tables
[i
].table
;
550 printk(" SMBIOS=0x%lx", config_tables
[i
].table
);
551 } else if (efi_guidcmp(config_tables
[i
].guid
, SAL_SYSTEM_TABLE_GUID
) == 0) {
552 efi
.sal_systab
= config_tables
[i
].table
;
553 printk(" SALsystab=0x%lx", config_tables
[i
].table
);
554 } else if (efi_guidcmp(config_tables
[i
].guid
, HCDP_TABLE_GUID
) == 0) {
555 efi
.hcdp
= config_tables
[i
].table
;
556 printk(" HCDP=0x%lx", config_tables
[i
].table
);
557 } else if (efi_guidcmp(config_tables
[i
].guid
,
558 PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID
) == 0) {
559 palo_phys
= config_tables
[i
].table
;
560 printk(" PALO=0x%lx", config_tables
[i
].table
);
565 if (palo_phys
!= EFI_INVALID_TABLE_ADDR
)
566 handle_palo(palo_phys
);
568 runtime
= __va(efi
.systab
->runtime
);
569 efi
.get_time
= phys_get_time
;
570 efi
.set_time
= phys_set_time
;
571 efi
.get_wakeup_time
= phys_get_wakeup_time
;
572 efi
.set_wakeup_time
= phys_set_wakeup_time
;
573 efi
.get_variable
= phys_get_variable
;
574 efi
.get_next_variable
= phys_get_next_variable
;
575 efi
.set_variable
= phys_set_variable
;
576 efi
.get_next_high_mono_count
= phys_get_next_high_mono_count
;
577 efi
.reset_system
= phys_reset_system
;
579 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
580 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
581 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
584 /* print EFI memory map: */
586 efi_memory_desc_t
*md
;
589 for (i
= 0, p
= efi_map_start
; p
< efi_map_end
;
590 ++i
, p
+= efi_desc_size
)
596 size
= md
->num_pages
<< EFI_PAGE_SHIFT
;
598 if ((size
>> 40) > 0) {
601 } else if ((size
>> 30) > 0) {
604 } else if ((size
>> 20) > 0) {
612 printk("mem%02d: type=%2u, attr=0x%016lx, "
613 "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
614 i
, md
->type
, md
->attribute
, md
->phys_addr
,
615 md
->phys_addr
+ efi_md_size(md
), size
, unit
);
621 efi_enter_virtual_mode();
625 efi_enter_virtual_mode (void)
627 void *efi_map_start
, *efi_map_end
, *p
;
628 efi_memory_desc_t
*md
;
632 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
633 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
634 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
636 for (p
= efi_map_start
; p
< efi_map_end
; p
+= efi_desc_size
) {
638 if (md
->attribute
& EFI_MEMORY_RUNTIME
) {
640 * Some descriptors have multiple bits set, so the
641 * order of the tests is relevant.
643 if (md
->attribute
& EFI_MEMORY_WB
) {
644 md
->virt_addr
= (u64
) __va(md
->phys_addr
);
645 } else if (md
->attribute
& EFI_MEMORY_UC
) {
646 md
->virt_addr
= (u64
) ioremap(md
->phys_addr
, 0);
647 } else if (md
->attribute
& EFI_MEMORY_WC
) {
649 md
->virt_addr
= ia64_remap(md
->phys_addr
,
657 printk(KERN_INFO
"EFI_MEMORY_WC mapping\n");
658 md
->virt_addr
= (u64
) ioremap(md
->phys_addr
, 0);
660 } else if (md
->attribute
& EFI_MEMORY_WT
) {
662 md
->virt_addr
= ia64_remap(md
->phys_addr
,
670 printk(KERN_INFO
"EFI_MEMORY_WT mapping\n");
671 md
->virt_addr
= (u64
) ioremap(md
->phys_addr
, 0);
677 status
= efi_call_phys(__va(runtime
->set_virtual_address_map
),
678 ia64_boot_param
->efi_memmap_size
,
680 ia64_boot_param
->efi_memdesc_version
,
681 ia64_boot_param
->efi_memmap
);
682 if (status
!= EFI_SUCCESS
) {
683 printk(KERN_WARNING
"warning: unable to switch EFI into "
684 "virtual mode (status=%lu)\n", status
);
689 * Now that EFI is in virtual mode, we call the EFI functions more
692 efi
.get_time
= virt_get_time
;
693 efi
.set_time
= virt_set_time
;
694 efi
.get_wakeup_time
= virt_get_wakeup_time
;
695 efi
.set_wakeup_time
= virt_set_wakeup_time
;
696 efi
.get_variable
= virt_get_variable
;
697 efi
.get_next_variable
= virt_get_next_variable
;
698 efi
.set_variable
= virt_set_variable
;
699 efi
.get_next_high_mono_count
= virt_get_next_high_mono_count
;
700 efi
.reset_system
= virt_reset_system
;
704 * Walk the EFI memory map looking for the I/O port range. There can only be
705 * one entry of this type, other I/O port ranges should be described via ACPI.
708 efi_get_iobase (void)
710 void *efi_map_start
, *efi_map_end
, *p
;
711 efi_memory_desc_t
*md
;
714 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
715 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
716 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
718 for (p
= efi_map_start
; p
< efi_map_end
; p
+= efi_desc_size
) {
720 if (md
->type
== EFI_MEMORY_MAPPED_IO_PORT_SPACE
) {
721 if (md
->attribute
& EFI_MEMORY_UC
)
722 return md
->phys_addr
;
728 static struct kern_memdesc
*
729 kern_memory_descriptor (unsigned long phys_addr
)
731 struct kern_memdesc
*md
;
733 for (md
= kern_memmap
; md
->start
!= ~0UL; md
++) {
734 if (phys_addr
- md
->start
< (md
->num_pages
<< EFI_PAGE_SHIFT
))
740 static efi_memory_desc_t
*
741 efi_memory_descriptor (unsigned long phys_addr
)
743 void *efi_map_start
, *efi_map_end
, *p
;
744 efi_memory_desc_t
*md
;
747 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
748 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
749 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
751 for (p
= efi_map_start
; p
< efi_map_end
; p
+= efi_desc_size
) {
754 if (phys_addr
- md
->phys_addr
< efi_md_size(md
))
761 efi_memmap_intersects (unsigned long phys_addr
, unsigned long size
)
763 void *efi_map_start
, *efi_map_end
, *p
;
764 efi_memory_desc_t
*md
;
768 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
769 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
770 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
772 end
= phys_addr
+ size
;
774 for (p
= efi_map_start
; p
< efi_map_end
; p
+= efi_desc_size
) {
776 if (md
->phys_addr
< end
&& efi_md_end(md
) > phys_addr
)
783 efi_mem_type (unsigned long phys_addr
)
785 efi_memory_desc_t
*md
= efi_memory_descriptor(phys_addr
);
793 efi_mem_attributes (unsigned long phys_addr
)
795 efi_memory_desc_t
*md
= efi_memory_descriptor(phys_addr
);
798 return md
->attribute
;
801 EXPORT_SYMBOL(efi_mem_attributes
);
804 efi_mem_attribute (unsigned long phys_addr
, unsigned long size
)
806 unsigned long end
= phys_addr
+ size
;
807 efi_memory_desc_t
*md
= efi_memory_descriptor(phys_addr
);
814 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
815 * the kernel that firmware needs this region mapped.
817 attr
= md
->attribute
& ~EFI_MEMORY_RUNTIME
;
819 unsigned long md_end
= efi_md_end(md
);
824 md
= efi_memory_descriptor(md_end
);
825 if (!md
|| (md
->attribute
& ~EFI_MEMORY_RUNTIME
) != attr
)
828 return 0; /* never reached */
832 kern_mem_attribute (unsigned long phys_addr
, unsigned long size
)
834 unsigned long end
= phys_addr
+ size
;
835 struct kern_memdesc
*md
;
839 * This is a hack for ioremap calls before we set up kern_memmap.
840 * Maybe we should do efi_memmap_init() earlier instead.
843 attr
= efi_mem_attribute(phys_addr
, size
);
844 if (attr
& EFI_MEMORY_WB
)
845 return EFI_MEMORY_WB
;
849 md
= kern_memory_descriptor(phys_addr
);
853 attr
= md
->attribute
;
855 unsigned long md_end
= kmd_end(md
);
860 md
= kern_memory_descriptor(md_end
);
861 if (!md
|| md
->attribute
!= attr
)
864 return 0; /* never reached */
866 EXPORT_SYMBOL(kern_mem_attribute
);
869 valid_phys_addr_range (unsigned long phys_addr
, unsigned long size
)
874 * /dev/mem reads and writes use copy_to_user(), which implicitly
875 * uses a granule-sized kernel identity mapping. It's really
876 * only safe to do this for regions in kern_memmap. For more
877 * details, see Documentation/ia64/aliasing.txt.
879 attr
= kern_mem_attribute(phys_addr
, size
);
880 if (attr
& EFI_MEMORY_WB
|| attr
& EFI_MEMORY_UC
)
886 valid_mmap_phys_addr_range (unsigned long pfn
, unsigned long size
)
888 unsigned long phys_addr
= pfn
<< PAGE_SHIFT
;
891 attr
= efi_mem_attribute(phys_addr
, size
);
894 * /dev/mem mmap uses normal user pages, so we don't need the entire
895 * granule, but the entire region we're mapping must support the same
898 if (attr
& EFI_MEMORY_WB
|| attr
& EFI_MEMORY_UC
)
902 * Intel firmware doesn't tell us about all the MMIO regions, so
903 * in general we have to allow mmap requests. But if EFI *does*
904 * tell us about anything inside this region, we should deny it.
905 * The user can always map a smaller region to avoid the overlap.
907 if (efi_memmap_intersects(phys_addr
, size
))
914 phys_mem_access_prot(struct file
*file
, unsigned long pfn
, unsigned long size
,
917 unsigned long phys_addr
= pfn
<< PAGE_SHIFT
;
921 * For /dev/mem mmap, we use user mappings, but if the region is
922 * in kern_memmap (and hence may be covered by a kernel mapping),
923 * we must use the same attribute as the kernel mapping.
925 attr
= kern_mem_attribute(phys_addr
, size
);
926 if (attr
& EFI_MEMORY_WB
)
927 return pgprot_cacheable(vma_prot
);
928 else if (attr
& EFI_MEMORY_UC
)
929 return pgprot_noncached(vma_prot
);
932 * Some chipsets don't support UC access to memory. If
933 * WB is supported, we prefer that.
935 if (efi_mem_attribute(phys_addr
, size
) & EFI_MEMORY_WB
)
936 return pgprot_cacheable(vma_prot
);
938 return pgprot_noncached(vma_prot
);
942 efi_uart_console_only(void)
945 char *s
, name
[] = "ConOut";
946 efi_guid_t guid
= EFI_GLOBAL_VARIABLE_GUID
;
947 efi_char16_t
*utf16
, name_utf16
[32];
948 unsigned char data
[1024];
949 unsigned long size
= sizeof(data
);
950 struct efi_generic_dev_path
*hdr
, *end_addr
;
953 /* Convert to UTF-16 */
957 *utf16
++ = *s
++ & 0x7f;
960 status
= efi
.get_variable(name_utf16
, &guid
, NULL
, &size
, data
);
961 if (status
!= EFI_SUCCESS
) {
962 printk(KERN_ERR
"No EFI %s variable?\n", name
);
966 hdr
= (struct efi_generic_dev_path
*) data
;
967 end_addr
= (struct efi_generic_dev_path
*) ((u8
*) data
+ size
);
968 while (hdr
< end_addr
) {
969 if (hdr
->type
== EFI_DEV_MSG
&&
970 hdr
->sub_type
== EFI_DEV_MSG_UART
)
972 else if (hdr
->type
== EFI_DEV_END_PATH
||
973 hdr
->type
== EFI_DEV_END_PATH2
) {
976 if (hdr
->sub_type
== EFI_DEV_END_ENTIRE
)
980 hdr
= (struct efi_generic_dev_path
*)((u8
*) hdr
+ hdr
->length
);
982 printk(KERN_ERR
"Malformed %s value\n", name
);
987 * Look for the first granule aligned memory descriptor memory
988 * that is big enough to hold EFI memory map. Make sure this
989 * descriptor is atleast granule sized so it does not get trimmed
991 struct kern_memdesc
*
992 find_memmap_space (void)
994 u64 contig_low
=0, contig_high
=0;
996 void *efi_map_start
, *efi_map_end
, *p
, *q
;
997 efi_memory_desc_t
*md
, *pmd
= NULL
, *check_md
;
998 u64 space_needed
, efi_desc_size
;
999 unsigned long total_mem
= 0;
1001 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
1002 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
1003 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
1006 * Worst case: we need 3 kernel descriptors for each efi descriptor
1007 * (if every entry has a WB part in the middle, and UC head and tail),
1008 * plus one for the end marker.
1010 space_needed
= sizeof(kern_memdesc_t
) *
1011 (3 * (ia64_boot_param
->efi_memmap_size
/efi_desc_size
) + 1);
1013 for (p
= efi_map_start
; p
< efi_map_end
; pmd
= md
, p
+= efi_desc_size
) {
1018 if (pmd
== NULL
|| !efi_wb(pmd
) ||
1019 efi_md_end(pmd
) != md
->phys_addr
) {
1020 contig_low
= GRANULEROUNDUP(md
->phys_addr
);
1021 contig_high
= efi_md_end(md
);
1022 for (q
= p
+ efi_desc_size
; q
< efi_map_end
;
1023 q
+= efi_desc_size
) {
1025 if (!efi_wb(check_md
))
1027 if (contig_high
!= check_md
->phys_addr
)
1029 contig_high
= efi_md_end(check_md
);
1031 contig_high
= GRANULEROUNDDOWN(contig_high
);
1033 if (!is_memory_available(md
) || md
->type
== EFI_LOADER_DATA
)
1036 /* Round ends inward to granule boundaries */
1037 as
= max(contig_low
, md
->phys_addr
);
1038 ae
= min(contig_high
, efi_md_end(md
));
1040 /* keep within max_addr= and min_addr= command line arg */
1041 as
= max(as
, min_addr
);
1042 ae
= min(ae
, max_addr
);
1046 /* avoid going over mem= command line arg */
1047 if (total_mem
+ (ae
- as
) > mem_limit
)
1048 ae
-= total_mem
+ (ae
- as
) - mem_limit
;
1053 if (ae
- as
> space_needed
)
1056 if (p
>= efi_map_end
)
1057 panic("Can't allocate space for kernel memory descriptors");
1063 * Walk the EFI memory map and gather all memory available for kernel
1064 * to use. We can allocate partial granules only if the unavailable
1065 * parts exist, and are WB.
1068 efi_memmap_init(unsigned long *s
, unsigned long *e
)
1070 struct kern_memdesc
*k
, *prev
= NULL
;
1071 u64 contig_low
=0, contig_high
=0;
1073 void *efi_map_start
, *efi_map_end
, *p
, *q
;
1074 efi_memory_desc_t
*md
, *pmd
= NULL
, *check_md
;
1076 unsigned long total_mem
= 0;
1078 k
= kern_memmap
= find_memmap_space();
1080 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
1081 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
1082 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
1084 for (p
= efi_map_start
; p
< efi_map_end
; pmd
= md
, p
+= efi_desc_size
) {
1088 (md
->type
== EFI_CONVENTIONAL_MEMORY
||
1089 md
->type
== EFI_BOOT_SERVICES_DATA
)) {
1090 k
->attribute
= EFI_MEMORY_UC
;
1091 k
->start
= md
->phys_addr
;
1092 k
->num_pages
= md
->num_pages
;
1097 if (pmd
== NULL
|| !efi_wb(pmd
) ||
1098 efi_md_end(pmd
) != md
->phys_addr
) {
1099 contig_low
= GRANULEROUNDUP(md
->phys_addr
);
1100 contig_high
= efi_md_end(md
);
1101 for (q
= p
+ efi_desc_size
; q
< efi_map_end
;
1102 q
+= efi_desc_size
) {
1104 if (!efi_wb(check_md
))
1106 if (contig_high
!= check_md
->phys_addr
)
1108 contig_high
= efi_md_end(check_md
);
1110 contig_high
= GRANULEROUNDDOWN(contig_high
);
1112 if (!is_memory_available(md
))
1115 #ifdef CONFIG_CRASH_DUMP
1116 /* saved_max_pfn should ignore max_addr= command line arg */
1117 if (saved_max_pfn
< (efi_md_end(md
) >> PAGE_SHIFT
))
1118 saved_max_pfn
= (efi_md_end(md
) >> PAGE_SHIFT
);
1121 * Round ends inward to granule boundaries
1122 * Give trimmings to uncached allocator
1124 if (md
->phys_addr
< contig_low
) {
1125 lim
= min(efi_md_end(md
), contig_low
);
1127 if (k
> kern_memmap
&&
1128 (k
-1)->attribute
== EFI_MEMORY_UC
&&
1129 kmd_end(k
-1) == md
->phys_addr
) {
1131 (lim
- md
->phys_addr
)
1134 k
->attribute
= EFI_MEMORY_UC
;
1135 k
->start
= md
->phys_addr
;
1136 k
->num_pages
= (lim
- md
->phys_addr
)
1145 if (efi_md_end(md
) > contig_high
) {
1146 lim
= max(md
->phys_addr
, contig_high
);
1148 if (lim
== md
->phys_addr
&& k
> kern_memmap
&&
1149 (k
-1)->attribute
== EFI_MEMORY_UC
&&
1150 kmd_end(k
-1) == md
->phys_addr
) {
1151 (k
-1)->num_pages
+= md
->num_pages
;
1153 k
->attribute
= EFI_MEMORY_UC
;
1155 k
->num_pages
= (efi_md_end(md
) - lim
)
1162 ae
= efi_md_end(md
);
1164 /* keep within max_addr= and min_addr= command line arg */
1165 as
= max(as
, min_addr
);
1166 ae
= min(ae
, max_addr
);
1170 /* avoid going over mem= command line arg */
1171 if (total_mem
+ (ae
- as
) > mem_limit
)
1172 ae
-= total_mem
+ (ae
- as
) - mem_limit
;
1176 if (prev
&& kmd_end(prev
) == md
->phys_addr
) {
1177 prev
->num_pages
+= (ae
- as
) >> EFI_PAGE_SHIFT
;
1178 total_mem
+= ae
- as
;
1181 k
->attribute
= EFI_MEMORY_WB
;
1183 k
->num_pages
= (ae
- as
) >> EFI_PAGE_SHIFT
;
1184 total_mem
+= ae
- as
;
1187 k
->start
= ~0L; /* end-marker */
1189 /* reserve the memory we are using for kern_memmap */
1190 *s
= (u64
)kern_memmap
;
1197 efi_initialize_iomem_resources(struct resource
*code_resource
,
1198 struct resource
*data_resource
,
1199 struct resource
*bss_resource
)
1201 struct resource
*res
;
1202 void *efi_map_start
, *efi_map_end
, *p
;
1203 efi_memory_desc_t
*md
;
1206 unsigned long flags
;
1208 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
1209 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
1210 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
1214 for (p
= efi_map_start
; p
< efi_map_end
; p
+= efi_desc_size
) {
1217 if (md
->num_pages
== 0) /* should not happen */
1220 flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
1223 case EFI_MEMORY_MAPPED_IO
:
1224 case EFI_MEMORY_MAPPED_IO_PORT_SPACE
:
1227 case EFI_LOADER_CODE
:
1228 case EFI_LOADER_DATA
:
1229 case EFI_BOOT_SERVICES_DATA
:
1230 case EFI_BOOT_SERVICES_CODE
:
1231 case EFI_CONVENTIONAL_MEMORY
:
1232 if (md
->attribute
& EFI_MEMORY_WP
) {
1233 name
= "System ROM";
1234 flags
|= IORESOURCE_READONLY
;
1236 name
= "System RAM";
1240 case EFI_ACPI_MEMORY_NVS
:
1241 name
= "ACPI Non-volatile Storage";
1244 case EFI_UNUSABLE_MEMORY
:
1246 flags
|= IORESOURCE_DISABLED
;
1249 case EFI_RESERVED_TYPE
:
1250 case EFI_RUNTIME_SERVICES_CODE
:
1251 case EFI_RUNTIME_SERVICES_DATA
:
1252 case EFI_ACPI_RECLAIM_MEMORY
:
1258 if ((res
= kzalloc(sizeof(struct resource
),
1259 GFP_KERNEL
)) == NULL
) {
1261 "failed to allocate resource for iomem\n");
1266 res
->start
= md
->phys_addr
;
1267 res
->end
= md
->phys_addr
+ efi_md_size(md
) - 1;
1270 if (insert_resource(&iomem_resource
, res
) < 0)
1274 * We don't know which region contains
1275 * kernel data so we try it repeatedly and
1276 * let the resource manager test it.
1278 insert_resource(res
, code_resource
);
1279 insert_resource(res
, data_resource
);
1280 insert_resource(res
, bss_resource
);
1282 insert_resource(res
, &efi_memmap_res
);
1283 insert_resource(res
, &boot_param_res
);
1284 if (crashk_res
.end
> crashk_res
.start
)
1285 insert_resource(res
, &crashk_res
);
1292 /* find a block of memory aligned to 64M exclude reserved regions
1293 rsvd_regions are sorted
1295 unsigned long __init
1296 kdump_find_rsvd_region (unsigned long size
, struct rsvd_region
*r
, int n
)
1300 u64 alignment
= 1UL << _PAGE_SIZE_64M
;
1301 void *efi_map_start
, *efi_map_end
, *p
;
1302 efi_memory_desc_t
*md
;
1305 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
1306 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
1307 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
1309 for (p
= efi_map_start
; p
< efi_map_end
; p
+= efi_desc_size
) {
1313 start
= ALIGN(md
->phys_addr
, alignment
);
1314 end
= efi_md_end(md
);
1315 for (i
= 0; i
< n
; i
++) {
1316 if (__pa(r
[i
].start
) >= start
&& __pa(r
[i
].end
) < end
) {
1317 if (__pa(r
[i
].start
) > start
+ size
)
1319 start
= ALIGN(__pa(r
[i
].end
), alignment
);
1321 __pa(r
[i
+1].start
) < start
+ size
)
1327 if (end
> start
+ size
)
1332 "Cannot reserve 0x%lx byte of memory for crashdump\n", size
);
1337 #ifdef CONFIG_PROC_VMCORE
1338 /* locate the size find a the descriptor at a certain address */
1339 unsigned long __init
1340 vmcore_find_descriptor_size (unsigned long address
)
1342 void *efi_map_start
, *efi_map_end
, *p
;
1343 efi_memory_desc_t
*md
;
1345 unsigned long ret
= 0;
1347 efi_map_start
= __va(ia64_boot_param
->efi_memmap
);
1348 efi_map_end
= efi_map_start
+ ia64_boot_param
->efi_memmap_size
;
1349 efi_desc_size
= ia64_boot_param
->efi_memdesc_size
;
1351 for (p
= efi_map_start
; p
< efi_map_end
; p
+= efi_desc_size
) {
1353 if (efi_wb(md
) && md
->type
== EFI_LOADER_DATA
1354 && md
->phys_addr
== address
) {
1355 ret
= efi_md_size(md
);
1361 printk(KERN_WARNING
"Cannot locate EFI vmcore descriptor\n");