2 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
3 * <benh@kernel.crashing.org>
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; either version
8 * 2 of the License, or (at your option) any later version.
11 #include <linux/module.h>
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/bootmem.h>
24 #include <linux/lmb.h>
26 #include <asm/pgtable.h>
27 #include <asm/system.h>
28 #include <asm/processor.h>
30 #include <asm/mmu_context.h>
32 #include <asm/machdep.h>
33 #include <asm/cputable.h>
34 #include <asm/sections.h>
35 #include <asm/firmware.h>
37 #include <asm/vdso_datapage.h>
44 #define DBG(fmt...) printk(fmt)
49 /* Max supported size for symbol names */
50 #define MAX_SYMNAME 64
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
__attribute__((__section__(".data.page_aligned")));
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
117 * Some infos carried around for each of them during parsing at
122 Elf32_Ehdr
*hdr
; /* ptr to ELF */
123 Elf32_Sym
*dynsym
; /* ptr to .dynsym section */
124 unsigned long dynsymsize
; /* size of .dynsym section */
125 char *dynstr
; /* ptr to .dynstr section */
126 unsigned long text
; /* offset of .text section in .so */
133 unsigned long dynsymsize
;
140 static void dump_one_vdso_page(struct page
*pg
, struct page
*upg
)
142 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg
) << PAGE_SHIFT
),
145 if (upg
&& !IS_ERR(upg
) /* && pg != upg*/) {
146 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg
)
154 static void dump_vdso_pages(struct vm_area_struct
* vma
)
158 if (!vma
|| test_thread_flag(TIF_32BIT
)) {
159 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase
);
160 for (i
=0; i
<vdso32_pages
; i
++) {
161 struct page
*pg
= virt_to_page(vdso32_kbase
+
163 struct page
*upg
= (vma
&& vma
->vm_mm
) ?
164 follow_page(vma
, vma
->vm_start
+ i
*PAGE_SIZE
, 0)
166 dump_one_vdso_page(pg
, upg
);
169 if (!vma
|| !test_thread_flag(TIF_32BIT
)) {
170 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase
);
171 for (i
=0; i
<vdso64_pages
; i
++) {
172 struct page
*pg
= virt_to_page(vdso64_kbase
+
174 struct page
*upg
= (vma
&& vma
->vm_mm
) ?
175 follow_page(vma
, vma
->vm_start
+ i
*PAGE_SIZE
, 0)
177 dump_one_vdso_page(pg
, upg
);
184 * This is called from binfmt_elf, we create the special vma for the
185 * vDSO and insert it into the mm struct tree
187 int arch_setup_additional_pages(struct linux_binprm
*bprm
, int uses_interp
)
189 struct mm_struct
*mm
= current
->mm
;
190 struct page
**vdso_pagelist
;
191 unsigned long vdso_pages
;
192 unsigned long vdso_base
;
199 if (test_thread_flag(TIF_32BIT
)) {
200 vdso_pagelist
= vdso32_pagelist
;
201 vdso_pages
= vdso32_pages
;
202 vdso_base
= VDSO32_MBASE
;
204 vdso_pagelist
= vdso64_pagelist
;
205 vdso_pages
= vdso64_pages
;
206 vdso_base
= VDSO64_MBASE
;
209 vdso_pagelist
= vdso32_pagelist
;
210 vdso_pages
= vdso32_pages
;
211 vdso_base
= VDSO32_MBASE
;
214 current
->mm
->context
.vdso_base
= 0;
216 /* vDSO has a problem and was disabled, just don't "enable" it for the
221 /* Add a page to the vdso size for the data page */
225 * pick a base address for the vDSO in process space. We try to put it
226 * at vdso_base which is the "natural" base for it, but we might fail
227 * and end up putting it elsewhere.
229 down_write(&mm
->mmap_sem
);
230 vdso_base
= get_unmapped_area(NULL
, vdso_base
,
231 vdso_pages
<< PAGE_SHIFT
, 0, 0);
232 if (IS_ERR_VALUE(vdso_base
)) {
238 * our vma flags don't have VM_WRITE so by default, the process isn't
239 * allowed to write those pages.
240 * gdb can break that with ptrace interface, and thus trigger COW on
241 * those pages but it's then your responsibility to never do that on
242 * the "data" page of the vDSO or you'll stop getting kernel updates
243 * and your nice userland gettimeofday will be totally dead.
244 * It's fine to use that for setting breakpoints in the vDSO code
247 * Make sure the vDSO gets into every core dump.
248 * Dumping its contents makes post-mortem fully interpretable later
249 * without matching up the same kernel and hardware config to see
250 * what PC values meant.
252 rc
= install_special_mapping(mm
, vdso_base
, vdso_pages
<< PAGE_SHIFT
,
254 VM_MAYREAD
|VM_MAYWRITE
|VM_MAYEXEC
|
260 /* Put vDSO base into mm struct */
261 current
->mm
->context
.vdso_base
= vdso_base
;
263 up_write(&mm
->mmap_sem
);
267 up_write(&mm
->mmap_sem
);
271 const char *arch_vma_name(struct vm_area_struct
*vma
)
273 if (vma
->vm_mm
&& vma
->vm_start
== vma
->vm_mm
->context
.vdso_base
)
280 static void * __init
find_section32(Elf32_Ehdr
*ehdr
, const char *secname
,
287 /* Grab section headers and strings so we can tell who is who */
288 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
289 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
291 /* Find the section they want */
292 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
293 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
295 *size
= sechdrs
[i
].sh_size
;
296 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
303 static Elf32_Sym
* __init
find_symbol32(struct lib32_elfinfo
*lib
,
307 char name
[MAX_SYMNAME
], *c
;
309 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf32_Sym
)); i
++) {
310 if (lib
->dynsym
[i
].st_name
== 0)
312 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
314 c
= strchr(name
, '@');
317 if (strcmp(symname
, name
) == 0)
318 return &lib
->dynsym
[i
];
323 /* Note that we assume the section is .text and the symbol is relative to
326 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
329 Elf32_Sym
*sym
= find_symbol32(lib
, symname
);
332 printk(KERN_WARNING
"vDSO32: function %s not found !\n",
336 return sym
->st_value
- VDSO32_LBASE
;
339 static int __init
vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
340 struct lib64_elfinfo
*v64
,
341 const char *orig
, const char *fix
)
343 Elf32_Sym
*sym32_gen
, *sym32_fix
;
345 sym32_gen
= find_symbol32(v32
, orig
);
346 if (sym32_gen
== NULL
) {
347 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", orig
);
351 sym32_gen
->st_name
= 0;
354 sym32_fix
= find_symbol32(v32
, fix
);
355 if (sym32_fix
== NULL
) {
356 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", fix
);
359 sym32_gen
->st_value
= sym32_fix
->st_value
;
360 sym32_gen
->st_size
= sym32_fix
->st_size
;
361 sym32_gen
->st_info
= sym32_fix
->st_info
;
362 sym32_gen
->st_other
= sym32_fix
->st_other
;
363 sym32_gen
->st_shndx
= sym32_fix
->st_shndx
;
371 static void * __init
find_section64(Elf64_Ehdr
*ehdr
, const char *secname
,
378 /* Grab section headers and strings so we can tell who is who */
379 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
380 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
382 /* Find the section they want */
383 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
384 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
386 *size
= sechdrs
[i
].sh_size
;
387 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
395 static Elf64_Sym
* __init
find_symbol64(struct lib64_elfinfo
*lib
,
399 char name
[MAX_SYMNAME
], *c
;
401 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf64_Sym
)); i
++) {
402 if (lib
->dynsym
[i
].st_name
== 0)
404 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
406 c
= strchr(name
, '@');
409 if (strcmp(symname
, name
) == 0)
410 return &lib
->dynsym
[i
];
415 /* Note that we assume the section is .text and the symbol is relative to
418 static unsigned long __init
find_function64(struct lib64_elfinfo
*lib
,
421 Elf64_Sym
*sym
= find_symbol64(lib
, symname
);
424 printk(KERN_WARNING
"vDSO64: function %s not found !\n",
428 #ifdef VDS64_HAS_DESCRIPTORS
429 return *((u64
*)(vdso64_kbase
+ sym
->st_value
- VDSO64_LBASE
)) -
432 return sym
->st_value
- VDSO64_LBASE
;
436 static int __init
vdso_do_func_patch64(struct lib32_elfinfo
*v32
,
437 struct lib64_elfinfo
*v64
,
438 const char *orig
, const char *fix
)
440 Elf64_Sym
*sym64_gen
, *sym64_fix
;
442 sym64_gen
= find_symbol64(v64
, orig
);
443 if (sym64_gen
== NULL
) {
444 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", orig
);
448 sym64_gen
->st_name
= 0;
451 sym64_fix
= find_symbol64(v64
, fix
);
452 if (sym64_fix
== NULL
) {
453 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", fix
);
456 sym64_gen
->st_value
= sym64_fix
->st_value
;
457 sym64_gen
->st_size
= sym64_fix
->st_size
;
458 sym64_gen
->st_info
= sym64_fix
->st_info
;
459 sym64_gen
->st_other
= sym64_fix
->st_other
;
460 sym64_gen
->st_shndx
= sym64_fix
->st_shndx
;
465 #endif /* CONFIG_PPC64 */
468 static __init
int vdso_do_find_sections(struct lib32_elfinfo
*v32
,
469 struct lib64_elfinfo
*v64
)
474 * Locate symbol tables & text section
477 v32
->dynsym
= find_section32(v32
->hdr
, ".dynsym", &v32
->dynsymsize
);
478 v32
->dynstr
= find_section32(v32
->hdr
, ".dynstr", NULL
);
479 if (v32
->dynsym
== NULL
|| v32
->dynstr
== NULL
) {
480 printk(KERN_ERR
"vDSO32: required symbol section not found\n");
483 sect
= find_section32(v32
->hdr
, ".text", NULL
);
485 printk(KERN_ERR
"vDSO32: the .text section was not found\n");
488 v32
->text
= sect
- vdso32_kbase
;
491 v64
->dynsym
= find_section64(v64
->hdr
, ".dynsym", &v64
->dynsymsize
);
492 v64
->dynstr
= find_section64(v64
->hdr
, ".dynstr", NULL
);
493 if (v64
->dynsym
== NULL
|| v64
->dynstr
== NULL
) {
494 printk(KERN_ERR
"vDSO64: required symbol section not found\n");
497 sect
= find_section64(v64
->hdr
, ".text", NULL
);
499 printk(KERN_ERR
"vDSO64: the .text section was not found\n");
502 v64
->text
= sect
- vdso64_kbase
;
503 #endif /* CONFIG_PPC64 */
508 static __init
void vdso_setup_trampolines(struct lib32_elfinfo
*v32
,
509 struct lib64_elfinfo
*v64
)
512 * Find signal trampolines
516 vdso64_rt_sigtramp
= find_function64(v64
, "__kernel_sigtramp_rt64");
518 vdso32_sigtramp
= find_function32(v32
, "__kernel_sigtramp32");
519 vdso32_rt_sigtramp
= find_function32(v32
, "__kernel_sigtramp_rt32");
522 static __init
int vdso_fixup_datapage(struct lib32_elfinfo
*v32
,
523 struct lib64_elfinfo
*v64
)
529 sym64
= find_symbol64(v64
, "__kernel_datapage_offset");
531 printk(KERN_ERR
"vDSO64: Can't find symbol "
532 "__kernel_datapage_offset !\n");
535 *((int *)(vdso64_kbase
+ sym64
->st_value
- VDSO64_LBASE
)) =
536 (vdso64_pages
<< PAGE_SHIFT
) -
537 (sym64
->st_value
- VDSO64_LBASE
);
538 #endif /* CONFIG_PPC64 */
540 sym32
= find_symbol32(v32
, "__kernel_datapage_offset");
542 printk(KERN_ERR
"vDSO32: Can't find symbol "
543 "__kernel_datapage_offset !\n");
546 *((int *)(vdso32_kbase
+ (sym32
->st_value
- VDSO32_LBASE
))) =
547 (vdso32_pages
<< PAGE_SHIFT
) -
548 (sym32
->st_value
- VDSO32_LBASE
);
554 static __init
int vdso_fixup_features(struct lib32_elfinfo
*v32
,
555 struct lib64_elfinfo
*v64
)
558 unsigned long size32
;
562 unsigned long size64
;
564 start64
= find_section64(v64
->hdr
, "__ftr_fixup", &size64
);
566 do_feature_fixups(cur_cpu_spec
->cpu_features
,
567 start64
, start64
+ size64
);
569 start64
= find_section64(v64
->hdr
, "__mmu_ftr_fixup", &size64
);
571 do_feature_fixups(cur_cpu_spec
->mmu_features
,
572 start64
, start64
+ size64
);
574 start64
= find_section64(v64
->hdr
, "__fw_ftr_fixup", &size64
);
576 do_feature_fixups(powerpc_firmware_features
,
577 start64
, start64
+ size64
);
579 start64
= find_section64(v64
->hdr
, "__lwsync_fixup", &size64
);
581 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
582 start64
, start64
+ size64
);
583 #endif /* CONFIG_PPC64 */
585 start32
= find_section32(v32
->hdr
, "__ftr_fixup", &size32
);
587 do_feature_fixups(cur_cpu_spec
->cpu_features
,
588 start32
, start32
+ size32
);
590 start32
= find_section32(v32
->hdr
, "__mmu_ftr_fixup", &size32
);
592 do_feature_fixups(cur_cpu_spec
->mmu_features
,
593 start32
, start32
+ size32
);
596 start32
= find_section32(v32
->hdr
, "__fw_ftr_fixup", &size32
);
598 do_feature_fixups(powerpc_firmware_features
,
599 start32
, start32
+ size32
);
600 #endif /* CONFIG_PPC64 */
602 start32
= find_section32(v32
->hdr
, "__lwsync_fixup", &size32
);
604 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
605 start32
, start32
+ size32
);
610 static __init
int vdso_fixup_alt_funcs(struct lib32_elfinfo
*v32
,
611 struct lib64_elfinfo
*v64
)
615 for (i
= 0; i
< ARRAY_SIZE(vdso_patches
); i
++) {
616 struct vdso_patch_def
*patch
= &vdso_patches
[i
];
617 int match
= (cur_cpu_spec
->cpu_features
& patch
->ftr_mask
)
622 DBG("replacing %s with %s...\n", patch
->gen_name
,
623 patch
->fix_name
? "NONE" : patch
->fix_name
);
626 * Patch the 32 bits and 64 bits symbols. Note that we do not
627 * patch the "." symbol on 64 bits.
628 * It would be easy to do, but doesn't seem to be necessary,
629 * patching the OPD symbol is enough.
631 vdso_do_func_patch32(v32
, v64
, patch
->gen_name
,
634 vdso_do_func_patch64(v32
, v64
, patch
->gen_name
,
636 #endif /* CONFIG_PPC64 */
643 static __init
int vdso_setup(void)
645 struct lib32_elfinfo v32
;
646 struct lib64_elfinfo v64
;
648 v32
.hdr
= vdso32_kbase
;
650 v64
.hdr
= vdso64_kbase
;
652 if (vdso_do_find_sections(&v32
, &v64
))
655 if (vdso_fixup_datapage(&v32
, &v64
))
658 if (vdso_fixup_features(&v32
, &v64
))
661 if (vdso_fixup_alt_funcs(&v32
, &v64
))
664 vdso_setup_trampolines(&v32
, &v64
);
670 * Called from setup_arch to initialize the bitmap of available
671 * syscalls in the systemcfg page
673 static void __init
vdso_setup_syscall_map(void)
676 extern unsigned long *sys_call_table
;
677 extern unsigned long sys_ni_syscall
;
680 for (i
= 0; i
< __NR_syscalls
; i
++) {
682 if (sys_call_table
[i
*2] != sys_ni_syscall
)
683 vdso_data
->syscall_map_64
[i
>> 5] |=
684 0x80000000UL
>> (i
& 0x1f);
685 if (sys_call_table
[i
*2+1] != sys_ni_syscall
)
686 vdso_data
->syscall_map_32
[i
>> 5] |=
687 0x80000000UL
>> (i
& 0x1f);
688 #else /* CONFIG_PPC64 */
689 if (sys_call_table
[i
] != sys_ni_syscall
)
690 vdso_data
->syscall_map_32
[i
>> 5] |=
691 0x80000000UL
>> (i
& 0x1f);
692 #endif /* CONFIG_PPC64 */
697 static int __init
vdso_init(void)
703 * Fill up the "systemcfg" stuff for backward compatiblity
705 strcpy((char *)vdso_data
->eye_catcher
, "SYSTEMCFG:PPC64");
706 vdso_data
->version
.major
= SYSTEMCFG_MAJOR
;
707 vdso_data
->version
.minor
= SYSTEMCFG_MINOR
;
708 vdso_data
->processor
= mfspr(SPRN_PVR
);
710 * Fake the old platform number for pSeries and iSeries and add
711 * in LPAR bit if necessary
713 vdso_data
->platform
= machine_is(iseries
) ? 0x200 : 0x100;
714 if (firmware_has_feature(FW_FEATURE_LPAR
))
715 vdso_data
->platform
|= 1;
716 vdso_data
->physicalMemorySize
= lmb_phys_mem_size();
717 vdso_data
->dcache_size
= ppc64_caches
.dsize
;
718 vdso_data
->dcache_line_size
= ppc64_caches
.dline_size
;
719 vdso_data
->icache_size
= ppc64_caches
.isize
;
720 vdso_data
->icache_line_size
= ppc64_caches
.iline_size
;
722 /* XXXOJN: Blocks should be added to ppc64_caches and used instead */
723 vdso_data
->dcache_block_size
= ppc64_caches
.dline_size
;
724 vdso_data
->icache_block_size
= ppc64_caches
.iline_size
;
725 vdso_data
->dcache_log_block_size
= ppc64_caches
.log_dline_size
;
726 vdso_data
->icache_log_block_size
= ppc64_caches
.log_iline_size
;
729 * Calculate the size of the 64 bits vDSO
731 vdso64_pages
= (&vdso64_end
- &vdso64_start
) >> PAGE_SHIFT
;
732 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase
, vdso64_pages
);
734 vdso_data
->dcache_block_size
= L1_CACHE_BYTES
;
735 vdso_data
->dcache_log_block_size
= L1_CACHE_SHIFT
;
736 vdso_data
->icache_block_size
= L1_CACHE_BYTES
;
737 vdso_data
->icache_log_block_size
= L1_CACHE_SHIFT
;
738 #endif /* CONFIG_PPC64 */
742 * Calculate the size of the 32 bits vDSO
744 vdso32_pages
= (&vdso32_end
- &vdso32_start
) >> PAGE_SHIFT
;
745 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase
, vdso32_pages
);
749 * Setup the syscall map in the vDOS
751 vdso_setup_syscall_map();
754 * Initialize the vDSO images in memory, that is do necessary
755 * fixups of vDSO symbols, locate trampolines, etc...
758 printk(KERN_ERR
"vDSO setup failure, not enabled !\n");
766 /* Make sure pages are in the correct state */
767 vdso32_pagelist
= kzalloc(sizeof(struct page
*) * (vdso32_pages
+ 2),
769 BUG_ON(vdso32_pagelist
== NULL
);
770 for (i
= 0; i
< vdso32_pages
; i
++) {
771 struct page
*pg
= virt_to_page(vdso32_kbase
+ i
*PAGE_SIZE
);
772 ClearPageReserved(pg
);
774 vdso32_pagelist
[i
] = pg
;
776 vdso32_pagelist
[i
++] = virt_to_page(vdso_data
);
777 vdso32_pagelist
[i
] = NULL
;
780 vdso64_pagelist
= kzalloc(sizeof(struct page
*) * (vdso64_pages
+ 2),
782 BUG_ON(vdso64_pagelist
== NULL
);
783 for (i
= 0; i
< vdso64_pages
; i
++) {
784 struct page
*pg
= virt_to_page(vdso64_kbase
+ i
*PAGE_SIZE
);
785 ClearPageReserved(pg
);
787 vdso64_pagelist
[i
] = pg
;
789 vdso64_pagelist
[i
++] = virt_to_page(vdso_data
);
790 vdso64_pagelist
[i
] = NULL
;
791 #endif /* CONFIG_PPC64 */
793 get_page(virt_to_page(vdso_data
));
800 arch_initcall(vdso_init
);
802 int in_gate_area_no_task(unsigned long addr
)
807 int in_gate_area(struct task_struct
*task
, unsigned long addr
)
812 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)