3 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
4 * <benh@kernel.crashing.org>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 #include <linux/errno.h>
13 #include <linux/sched.h>
14 #include <linux/kernel.h>
16 #include <linux/smp.h>
17 #include <linux/stddef.h>
18 #include <linux/unistd.h>
19 #include <linux/slab.h>
20 #include <linux/user.h>
21 #include <linux/elf.h>
22 #include <linux/security.h>
23 #include <linux/memblock.h>
25 #include <asm/pgtable.h>
26 #include <asm/processor.h>
28 #include <asm/mmu_context.h>
30 #include <asm/machdep.h>
31 #include <asm/cputable.h>
32 #include <asm/sections.h>
33 #include <asm/firmware.h>
35 #include <asm/vdso_datapage.h>
36 #include <asm/setup.h>
41 #define DBG(fmt...) printk(fmt)
46 /* Max supported size for symbol names */
47 #define MAX_SYMNAME 64
49 /* The alignment of the vDSO */
50 #define VDSO_ALIGNMENT (1 << 16)
52 extern char vdso32_start
, vdso32_end
;
53 static void *vdso32_kbase
= &vdso32_start
;
54 static unsigned int vdso32_pages
;
55 static struct page
**vdso32_pagelist
;
56 unsigned long vdso32_sigtramp
;
57 unsigned long vdso32_rt_sigtramp
;
60 extern char vdso64_start
, vdso64_end
;
61 static void *vdso64_kbase
= &vdso64_start
;
62 static unsigned int vdso64_pages
;
63 static struct page
**vdso64_pagelist
;
64 unsigned long vdso64_rt_sigtramp
;
65 #endif /* CONFIG_PPC64 */
67 static int vdso_ready
;
70 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
71 * Once the early boot kernel code no longer needs to muck around
72 * with it, it will become dynamically allocated
75 struct vdso_data data
;
77 } vdso_data_store __page_aligned_data
;
78 struct vdso_data
*vdso_data
= &vdso_data_store
.data
;
80 /* Format of the patch table */
83 unsigned long ftr_mask
, ftr_value
;
88 /* Table of functions to patch based on the CPU type/revision
90 * Currently, we only change sync_dicache to do nothing on processors
91 * with a coherent icache
93 static struct vdso_patch_def vdso_patches
[] = {
95 CPU_FTR_COHERENT_ICACHE
, CPU_FTR_COHERENT_ICACHE
,
96 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
100 "__kernel_gettimeofday", NULL
104 "__kernel_clock_gettime", NULL
108 "__kernel_clock_getres", NULL
112 "__kernel_get_tbfreq", NULL
116 "__kernel_time", NULL
121 * Some infos carried around for each of them during parsing at
126 Elf32_Ehdr
*hdr
; /* ptr to ELF */
127 Elf32_Sym
*dynsym
; /* ptr to .dynsym section */
128 unsigned long dynsymsize
; /* size of .dynsym section */
129 char *dynstr
; /* ptr to .dynstr section */
130 unsigned long text
; /* offset of .text section in .so */
137 unsigned long dynsymsize
;
144 * This is called from binfmt_elf, we create the special vma for the
145 * vDSO and insert it into the mm struct tree
147 int arch_setup_additional_pages(struct linux_binprm
*bprm
, int uses_interp
)
149 struct mm_struct
*mm
= current
->mm
;
150 struct page
**vdso_pagelist
;
151 unsigned long vdso_pages
;
152 unsigned long vdso_base
;
159 if (is_32bit_task()) {
160 vdso_pagelist
= vdso32_pagelist
;
161 vdso_pages
= vdso32_pages
;
162 vdso_base
= VDSO32_MBASE
;
164 vdso_pagelist
= vdso64_pagelist
;
165 vdso_pages
= vdso64_pages
;
167 * On 64bit we don't have a preferred map address. This
168 * allows get_unmapped_area to find an area near other mmaps
169 * and most likely share a SLB entry.
174 vdso_pagelist
= vdso32_pagelist
;
175 vdso_pages
= vdso32_pages
;
176 vdso_base
= VDSO32_MBASE
;
179 current
->mm
->context
.vdso_base
= 0;
181 /* vDSO has a problem and was disabled, just don't "enable" it for the
186 /* Add a page to the vdso size for the data page */
190 * pick a base address for the vDSO in process space. We try to put it
191 * at vdso_base which is the "natural" base for it, but we might fail
192 * and end up putting it elsewhere.
193 * Add enough to the size so that the result can be aligned.
195 down_write(&mm
->mmap_sem
);
196 vdso_base
= get_unmapped_area(NULL
, vdso_base
,
197 (vdso_pages
<< PAGE_SHIFT
) +
198 ((VDSO_ALIGNMENT
- 1) & PAGE_MASK
),
200 if (IS_ERR_VALUE(vdso_base
)) {
205 /* Add required alignment. */
206 vdso_base
= ALIGN(vdso_base
, VDSO_ALIGNMENT
);
209 * Put vDSO base into mm struct. We need to do this before calling
210 * install_special_mapping or the perf counter mmap tracking code
211 * will fail to recognise it as a vDSO (since arch_vma_name fails).
213 current
->mm
->context
.vdso_base
= vdso_base
;
216 * our vma flags don't have VM_WRITE so by default, the process isn't
217 * allowed to write those pages.
218 * gdb can break that with ptrace interface, and thus trigger COW on
219 * those pages but it's then your responsibility to never do that on
220 * the "data" page of the vDSO or you'll stop getting kernel updates
221 * and your nice userland gettimeofday will be totally dead.
222 * It's fine to use that for setting breakpoints in the vDSO code
225 rc
= install_special_mapping(mm
, vdso_base
, vdso_pages
<< PAGE_SHIFT
,
227 VM_MAYREAD
|VM_MAYWRITE
|VM_MAYEXEC
,
230 current
->mm
->context
.vdso_base
= 0;
234 up_write(&mm
->mmap_sem
);
238 up_write(&mm
->mmap_sem
);
242 const char *arch_vma_name(struct vm_area_struct
*vma
)
244 if (vma
->vm_mm
&& vma
->vm_start
== vma
->vm_mm
->context
.vdso_base
)
251 static void * __init
find_section32(Elf32_Ehdr
*ehdr
, const char *secname
,
258 /* Grab section headers and strings so we can tell who is who */
259 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
260 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
262 /* Find the section they want */
263 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
264 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
266 *size
= sechdrs
[i
].sh_size
;
267 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
274 static Elf32_Sym
* __init
find_symbol32(struct lib32_elfinfo
*lib
,
278 char name
[MAX_SYMNAME
], *c
;
280 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf32_Sym
)); i
++) {
281 if (lib
->dynsym
[i
].st_name
== 0)
283 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
285 c
= strchr(name
, '@');
288 if (strcmp(symname
, name
) == 0)
289 return &lib
->dynsym
[i
];
294 /* Note that we assume the section is .text and the symbol is relative to
297 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
300 Elf32_Sym
*sym
= find_symbol32(lib
, symname
);
303 printk(KERN_WARNING
"vDSO32: function %s not found !\n",
307 return sym
->st_value
- VDSO32_LBASE
;
310 static int __init
vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
311 struct lib64_elfinfo
*v64
,
312 const char *orig
, const char *fix
)
314 Elf32_Sym
*sym32_gen
, *sym32_fix
;
316 sym32_gen
= find_symbol32(v32
, orig
);
317 if (sym32_gen
== NULL
) {
318 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", orig
);
322 sym32_gen
->st_name
= 0;
325 sym32_fix
= find_symbol32(v32
, fix
);
326 if (sym32_fix
== NULL
) {
327 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", fix
);
330 sym32_gen
->st_value
= sym32_fix
->st_value
;
331 sym32_gen
->st_size
= sym32_fix
->st_size
;
332 sym32_gen
->st_info
= sym32_fix
->st_info
;
333 sym32_gen
->st_other
= sym32_fix
->st_other
;
334 sym32_gen
->st_shndx
= sym32_fix
->st_shndx
;
342 static void * __init
find_section64(Elf64_Ehdr
*ehdr
, const char *secname
,
349 /* Grab section headers and strings so we can tell who is who */
350 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
351 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
353 /* Find the section they want */
354 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
355 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
357 *size
= sechdrs
[i
].sh_size
;
358 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
366 static Elf64_Sym
* __init
find_symbol64(struct lib64_elfinfo
*lib
,
370 char name
[MAX_SYMNAME
], *c
;
372 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf64_Sym
)); i
++) {
373 if (lib
->dynsym
[i
].st_name
== 0)
375 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
377 c
= strchr(name
, '@');
380 if (strcmp(symname
, name
) == 0)
381 return &lib
->dynsym
[i
];
386 /* Note that we assume the section is .text and the symbol is relative to
389 static unsigned long __init
find_function64(struct lib64_elfinfo
*lib
,
392 Elf64_Sym
*sym
= find_symbol64(lib
, symname
);
395 printk(KERN_WARNING
"vDSO64: function %s not found !\n",
399 #ifdef VDS64_HAS_DESCRIPTORS
400 return *((u64
*)(vdso64_kbase
+ sym
->st_value
- VDSO64_LBASE
)) -
403 return sym
->st_value
- VDSO64_LBASE
;
407 static int __init
vdso_do_func_patch64(struct lib32_elfinfo
*v32
,
408 struct lib64_elfinfo
*v64
,
409 const char *orig
, const char *fix
)
411 Elf64_Sym
*sym64_gen
, *sym64_fix
;
413 sym64_gen
= find_symbol64(v64
, orig
);
414 if (sym64_gen
== NULL
) {
415 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", orig
);
419 sym64_gen
->st_name
= 0;
422 sym64_fix
= find_symbol64(v64
, fix
);
423 if (sym64_fix
== NULL
) {
424 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", fix
);
427 sym64_gen
->st_value
= sym64_fix
->st_value
;
428 sym64_gen
->st_size
= sym64_fix
->st_size
;
429 sym64_gen
->st_info
= sym64_fix
->st_info
;
430 sym64_gen
->st_other
= sym64_fix
->st_other
;
431 sym64_gen
->st_shndx
= sym64_fix
->st_shndx
;
436 #endif /* CONFIG_PPC64 */
439 static __init
int vdso_do_find_sections(struct lib32_elfinfo
*v32
,
440 struct lib64_elfinfo
*v64
)
445 * Locate symbol tables & text section
448 v32
->dynsym
= find_section32(v32
->hdr
, ".dynsym", &v32
->dynsymsize
);
449 v32
->dynstr
= find_section32(v32
->hdr
, ".dynstr", NULL
);
450 if (v32
->dynsym
== NULL
|| v32
->dynstr
== NULL
) {
451 printk(KERN_ERR
"vDSO32: required symbol section not found\n");
454 sect
= find_section32(v32
->hdr
, ".text", NULL
);
456 printk(KERN_ERR
"vDSO32: the .text section was not found\n");
459 v32
->text
= sect
- vdso32_kbase
;
462 v64
->dynsym
= find_section64(v64
->hdr
, ".dynsym", &v64
->dynsymsize
);
463 v64
->dynstr
= find_section64(v64
->hdr
, ".dynstr", NULL
);
464 if (v64
->dynsym
== NULL
|| v64
->dynstr
== NULL
) {
465 printk(KERN_ERR
"vDSO64: required symbol section not found\n");
468 sect
= find_section64(v64
->hdr
, ".text", NULL
);
470 printk(KERN_ERR
"vDSO64: the .text section was not found\n");
473 v64
->text
= sect
- vdso64_kbase
;
474 #endif /* CONFIG_PPC64 */
479 static __init
void vdso_setup_trampolines(struct lib32_elfinfo
*v32
,
480 struct lib64_elfinfo
*v64
)
483 * Find signal trampolines
487 vdso64_rt_sigtramp
= find_function64(v64
, "__kernel_sigtramp_rt64");
489 vdso32_sigtramp
= find_function32(v32
, "__kernel_sigtramp32");
490 vdso32_rt_sigtramp
= find_function32(v32
, "__kernel_sigtramp_rt32");
493 static __init
int vdso_fixup_datapage(struct lib32_elfinfo
*v32
,
494 struct lib64_elfinfo
*v64
)
500 sym64
= find_symbol64(v64
, "__kernel_datapage_offset");
502 printk(KERN_ERR
"vDSO64: Can't find symbol "
503 "__kernel_datapage_offset !\n");
506 *((int *)(vdso64_kbase
+ sym64
->st_value
- VDSO64_LBASE
)) =
507 (vdso64_pages
<< PAGE_SHIFT
) -
508 (sym64
->st_value
- VDSO64_LBASE
);
509 #endif /* CONFIG_PPC64 */
511 sym32
= find_symbol32(v32
, "__kernel_datapage_offset");
513 printk(KERN_ERR
"vDSO32: Can't find symbol "
514 "__kernel_datapage_offset !\n");
517 *((int *)(vdso32_kbase
+ (sym32
->st_value
- VDSO32_LBASE
))) =
518 (vdso32_pages
<< PAGE_SHIFT
) -
519 (sym32
->st_value
- VDSO32_LBASE
);
525 static __init
int vdso_fixup_features(struct lib32_elfinfo
*v32
,
526 struct lib64_elfinfo
*v64
)
529 unsigned long size32
;
533 unsigned long size64
;
535 start64
= find_section64(v64
->hdr
, "__ftr_fixup", &size64
);
537 do_feature_fixups(cur_cpu_spec
->cpu_features
,
538 start64
, start64
+ size64
);
540 start64
= find_section64(v64
->hdr
, "__mmu_ftr_fixup", &size64
);
542 do_feature_fixups(cur_cpu_spec
->mmu_features
,
543 start64
, start64
+ size64
);
545 start64
= find_section64(v64
->hdr
, "__fw_ftr_fixup", &size64
);
547 do_feature_fixups(powerpc_firmware_features
,
548 start64
, start64
+ size64
);
550 start64
= find_section64(v64
->hdr
, "__lwsync_fixup", &size64
);
552 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
553 start64
, start64
+ size64
);
554 #endif /* CONFIG_PPC64 */
556 start32
= find_section32(v32
->hdr
, "__ftr_fixup", &size32
);
558 do_feature_fixups(cur_cpu_spec
->cpu_features
,
559 start32
, start32
+ size32
);
561 start32
= find_section32(v32
->hdr
, "__mmu_ftr_fixup", &size32
);
563 do_feature_fixups(cur_cpu_spec
->mmu_features
,
564 start32
, start32
+ size32
);
567 start32
= find_section32(v32
->hdr
, "__fw_ftr_fixup", &size32
);
569 do_feature_fixups(powerpc_firmware_features
,
570 start32
, start32
+ size32
);
571 #endif /* CONFIG_PPC64 */
573 start32
= find_section32(v32
->hdr
, "__lwsync_fixup", &size32
);
575 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
576 start32
, start32
+ size32
);
581 static __init
int vdso_fixup_alt_funcs(struct lib32_elfinfo
*v32
,
582 struct lib64_elfinfo
*v64
)
586 for (i
= 0; i
< ARRAY_SIZE(vdso_patches
); i
++) {
587 struct vdso_patch_def
*patch
= &vdso_patches
[i
];
588 int match
= (cur_cpu_spec
->cpu_features
& patch
->ftr_mask
)
593 DBG("replacing %s with %s...\n", patch
->gen_name
,
594 patch
->fix_name
? "NONE" : patch
->fix_name
);
597 * Patch the 32 bits and 64 bits symbols. Note that we do not
598 * patch the "." symbol on 64 bits.
599 * It would be easy to do, but doesn't seem to be necessary,
600 * patching the OPD symbol is enough.
602 vdso_do_func_patch32(v32
, v64
, patch
->gen_name
,
605 vdso_do_func_patch64(v32
, v64
, patch
->gen_name
,
607 #endif /* CONFIG_PPC64 */
614 static __init
int vdso_setup(void)
616 struct lib32_elfinfo v32
;
617 struct lib64_elfinfo v64
;
619 v32
.hdr
= vdso32_kbase
;
621 v64
.hdr
= vdso64_kbase
;
623 if (vdso_do_find_sections(&v32
, &v64
))
626 if (vdso_fixup_datapage(&v32
, &v64
))
629 if (vdso_fixup_features(&v32
, &v64
))
632 if (vdso_fixup_alt_funcs(&v32
, &v64
))
635 vdso_setup_trampolines(&v32
, &v64
);
641 * Called from setup_arch to initialize the bitmap of available
642 * syscalls in the systemcfg page
644 static void __init
vdso_setup_syscall_map(void)
647 extern unsigned long *sys_call_table
;
648 extern unsigned long sys_ni_syscall
;
651 for (i
= 0; i
< __NR_syscalls
; i
++) {
653 if (sys_call_table
[i
*2] != sys_ni_syscall
)
654 vdso_data
->syscall_map_64
[i
>> 5] |=
655 0x80000000UL
>> (i
& 0x1f);
656 if (sys_call_table
[i
*2+1] != sys_ni_syscall
)
657 vdso_data
->syscall_map_32
[i
>> 5] |=
658 0x80000000UL
>> (i
& 0x1f);
659 #else /* CONFIG_PPC64 */
660 if (sys_call_table
[i
] != sys_ni_syscall
)
661 vdso_data
->syscall_map_32
[i
>> 5] |=
662 0x80000000UL
>> (i
& 0x1f);
663 #endif /* CONFIG_PPC64 */
668 int vdso_getcpu_init(void)
670 unsigned long cpu
, node
, val
;
673 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
674 * in the next 16 bits. The VDSO uses this to implement getcpu().
677 WARN_ON_ONCE(cpu
> 0xffff);
679 node
= cpu_to_node(cpu
);
680 WARN_ON_ONCE(node
> 0xffff);
682 val
= (cpu
& 0xfff) | ((node
& 0xffff) << 16);
683 mtspr(SPRN_SPRG_VDSO_WRITE
, val
);
684 get_paca()->sprg_vdso
= val
;
690 /* We need to call this before SMP init */
691 early_initcall(vdso_getcpu_init
);
694 static int __init
vdso_init(void)
700 * Fill up the "systemcfg" stuff for backward compatibility
702 strcpy((char *)vdso_data
->eye_catcher
, "SYSTEMCFG:PPC64");
703 vdso_data
->version
.major
= SYSTEMCFG_MAJOR
;
704 vdso_data
->version
.minor
= SYSTEMCFG_MINOR
;
705 vdso_data
->processor
= mfspr(SPRN_PVR
);
707 * Fake the old platform number for pSeries and add
708 * in LPAR bit if necessary
710 vdso_data
->platform
= 0x100;
711 if (firmware_has_feature(FW_FEATURE_LPAR
))
712 vdso_data
->platform
|= 1;
713 vdso_data
->physicalMemorySize
= memblock_phys_mem_size();
714 vdso_data
->dcache_size
= ppc64_caches
.dsize
;
715 vdso_data
->dcache_line_size
= ppc64_caches
.dline_size
;
716 vdso_data
->icache_size
= ppc64_caches
.isize
;
717 vdso_data
->icache_line_size
= ppc64_caches
.iline_size
;
719 /* XXXOJN: Blocks should be added to ppc64_caches and used instead */
720 vdso_data
->dcache_block_size
= ppc64_caches
.dline_size
;
721 vdso_data
->icache_block_size
= ppc64_caches
.iline_size
;
722 vdso_data
->dcache_log_block_size
= ppc64_caches
.log_dline_size
;
723 vdso_data
->icache_log_block_size
= ppc64_caches
.log_iline_size
;
726 * Calculate the size of the 64 bits vDSO
728 vdso64_pages
= (&vdso64_end
- &vdso64_start
) >> PAGE_SHIFT
;
729 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase
, vdso64_pages
);
731 vdso_data
->dcache_block_size
= L1_CACHE_BYTES
;
732 vdso_data
->dcache_log_block_size
= L1_CACHE_SHIFT
;
733 vdso_data
->icache_block_size
= L1_CACHE_BYTES
;
734 vdso_data
->icache_log_block_size
= L1_CACHE_SHIFT
;
735 #endif /* CONFIG_PPC64 */
739 * Calculate the size of the 32 bits vDSO
741 vdso32_pages
= (&vdso32_end
- &vdso32_start
) >> PAGE_SHIFT
;
742 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase
, vdso32_pages
);
746 * Setup the syscall map in the vDOS
748 vdso_setup_syscall_map();
751 * Initialize the vDSO images in memory, that is do necessary
752 * fixups of vDSO symbols, locate trampolines, etc...
755 printk(KERN_ERR
"vDSO setup failure, not enabled !\n");
763 /* Make sure pages are in the correct state */
764 vdso32_pagelist
= kzalloc(sizeof(struct page
*) * (vdso32_pages
+ 2),
766 BUG_ON(vdso32_pagelist
== NULL
);
767 for (i
= 0; i
< vdso32_pages
; i
++) {
768 struct page
*pg
= virt_to_page(vdso32_kbase
+ i
*PAGE_SIZE
);
769 ClearPageReserved(pg
);
771 vdso32_pagelist
[i
] = pg
;
773 vdso32_pagelist
[i
++] = virt_to_page(vdso_data
);
774 vdso32_pagelist
[i
] = NULL
;
777 vdso64_pagelist
= kzalloc(sizeof(struct page
*) * (vdso64_pages
+ 2),
779 BUG_ON(vdso64_pagelist
== NULL
);
780 for (i
= 0; i
< vdso64_pages
; i
++) {
781 struct page
*pg
= virt_to_page(vdso64_kbase
+ i
*PAGE_SIZE
);
782 ClearPageReserved(pg
);
784 vdso64_pagelist
[i
] = pg
;
786 vdso64_pagelist
[i
++] = virt_to_page(vdso_data
);
787 vdso64_pagelist
[i
] = NULL
;
788 #endif /* CONFIG_PPC64 */
790 get_page(virt_to_page(vdso_data
));
797 arch_initcall(vdso_init
);