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/smp_lock.h>
18 #include <linux/stddef.h>
19 #include <linux/unistd.h>
20 #include <linux/slab.h>
21 #include <linux/user.h>
22 #include <linux/elf.h>
23 #include <linux/security.h>
24 #include <linux/bootmem.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
105 * Some infos carried around for each of them during parsing at
110 Elf32_Ehdr
*hdr
; /* ptr to ELF */
111 Elf32_Sym
*dynsym
; /* ptr to .dynsym section */
112 unsigned long dynsymsize
; /* size of .dynsym section */
113 char *dynstr
; /* ptr to .dynstr section */
114 unsigned long text
; /* offset of .text section in .so */
121 unsigned long dynsymsize
;
128 static void dump_one_vdso_page(struct page
*pg
, struct page
*upg
)
130 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg
) << PAGE_SHIFT
),
133 if (upg
/* && pg != upg*/) {
134 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg
)
142 static void dump_vdso_pages(struct vm_area_struct
* vma
)
146 if (!vma
|| test_thread_flag(TIF_32BIT
)) {
147 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase
);
148 for (i
=0; i
<vdso32_pages
; i
++) {
149 struct page
*pg
= virt_to_page(vdso32_kbase
+
151 struct page
*upg
= (vma
&& vma
->vm_mm
) ?
152 follow_page(vma
, vma
->vm_start
+ i
*PAGE_SIZE
, 0)
154 dump_one_vdso_page(pg
, upg
);
157 if (!vma
|| !test_thread_flag(TIF_32BIT
)) {
158 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase
);
159 for (i
=0; i
<vdso64_pages
; i
++) {
160 struct page
*pg
= virt_to_page(vdso64_kbase
+
162 struct page
*upg
= (vma
&& vma
->vm_mm
) ?
163 follow_page(vma
, vma
->vm_start
+ i
*PAGE_SIZE
, 0)
165 dump_one_vdso_page(pg
, upg
);
172 * This is called from binfmt_elf, we create the special vma for the
173 * vDSO and insert it into the mm struct tree
175 int arch_setup_additional_pages(struct linux_binprm
*bprm
,
176 int executable_stack
)
178 struct mm_struct
*mm
= current
->mm
;
179 struct page
**vdso_pagelist
;
180 unsigned long vdso_pages
;
181 unsigned long vdso_base
;
188 if (test_thread_flag(TIF_32BIT
)) {
189 vdso_pagelist
= vdso32_pagelist
;
190 vdso_pages
= vdso32_pages
;
191 vdso_base
= VDSO32_MBASE
;
193 vdso_pagelist
= vdso64_pagelist
;
194 vdso_pages
= vdso64_pages
;
195 vdso_base
= VDSO64_MBASE
;
198 vdso_pagelist
= vdso32_pagelist
;
199 vdso_pages
= vdso32_pages
;
200 vdso_base
= VDSO32_MBASE
;
203 current
->mm
->context
.vdso_base
= 0;
205 /* vDSO has a problem and was disabled, just don't "enable" it for the
210 /* Add a page to the vdso size for the data page */
214 * pick a base address for the vDSO in process space. We try to put it
215 * at vdso_base which is the "natural" base for it, but we might fail
216 * and end up putting it elsewhere.
218 down_write(&mm
->mmap_sem
);
219 vdso_base
= get_unmapped_area(NULL
, vdso_base
,
220 vdso_pages
<< PAGE_SHIFT
, 0, 0);
221 if (IS_ERR_VALUE(vdso_base
)) {
227 * our vma flags don't have VM_WRITE so by default, the process isn't
228 * allowed to write those pages.
229 * gdb can break that with ptrace interface, and thus trigger COW on
230 * those pages but it's then your responsibility to never do that on
231 * the "data" page of the vDSO or you'll stop getting kernel updates
232 * and your nice userland gettimeofday will be totally dead.
233 * It's fine to use that for setting breakpoints in the vDSO code
236 * Make sure the vDSO gets into every core dump.
237 * Dumping its contents makes post-mortem fully interpretable later
238 * without matching up the same kernel and hardware config to see
239 * what PC values meant.
241 rc
= install_special_mapping(mm
, vdso_base
, vdso_pages
<< PAGE_SHIFT
,
243 VM_MAYREAD
|VM_MAYWRITE
|VM_MAYEXEC
|
249 /* Put vDSO base into mm struct */
250 current
->mm
->context
.vdso_base
= vdso_base
;
252 up_write(&mm
->mmap_sem
);
256 up_write(&mm
->mmap_sem
);
260 const char *arch_vma_name(struct vm_area_struct
*vma
)
262 if (vma
->vm_mm
&& vma
->vm_start
== vma
->vm_mm
->context
.vdso_base
)
269 static void * __init
find_section32(Elf32_Ehdr
*ehdr
, const char *secname
,
276 /* Grab section headers and strings so we can tell who is who */
277 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
278 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
280 /* Find the section they want */
281 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
282 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
284 *size
= sechdrs
[i
].sh_size
;
285 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
292 static Elf32_Sym
* __init
find_symbol32(struct lib32_elfinfo
*lib
,
296 char name
[MAX_SYMNAME
], *c
;
298 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf32_Sym
)); i
++) {
299 if (lib
->dynsym
[i
].st_name
== 0)
301 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
303 c
= strchr(name
, '@');
306 if (strcmp(symname
, name
) == 0)
307 return &lib
->dynsym
[i
];
312 /* Note that we assume the section is .text and the symbol is relative to
315 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
318 Elf32_Sym
*sym
= find_symbol32(lib
, symname
);
321 printk(KERN_WARNING
"vDSO32: function %s not found !\n",
325 return sym
->st_value
- VDSO32_LBASE
;
328 static int vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
329 struct lib64_elfinfo
*v64
,
330 const char *orig
, const char *fix
)
332 Elf32_Sym
*sym32_gen
, *sym32_fix
;
334 sym32_gen
= find_symbol32(v32
, orig
);
335 if (sym32_gen
== NULL
) {
336 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", orig
);
340 sym32_gen
->st_name
= 0;
343 sym32_fix
= find_symbol32(v32
, fix
);
344 if (sym32_fix
== NULL
) {
345 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", fix
);
348 sym32_gen
->st_value
= sym32_fix
->st_value
;
349 sym32_gen
->st_size
= sym32_fix
->st_size
;
350 sym32_gen
->st_info
= sym32_fix
->st_info
;
351 sym32_gen
->st_other
= sym32_fix
->st_other
;
352 sym32_gen
->st_shndx
= sym32_fix
->st_shndx
;
360 static void * __init
find_section64(Elf64_Ehdr
*ehdr
, const char *secname
,
367 /* Grab section headers and strings so we can tell who is who */
368 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
369 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
371 /* Find the section they want */
372 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
373 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
375 *size
= sechdrs
[i
].sh_size
;
376 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
384 static Elf64_Sym
* __init
find_symbol64(struct lib64_elfinfo
*lib
,
388 char name
[MAX_SYMNAME
], *c
;
390 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf64_Sym
)); i
++) {
391 if (lib
->dynsym
[i
].st_name
== 0)
393 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
395 c
= strchr(name
, '@');
398 if (strcmp(symname
, name
) == 0)
399 return &lib
->dynsym
[i
];
404 /* Note that we assume the section is .text and the symbol is relative to
407 static unsigned long __init
find_function64(struct lib64_elfinfo
*lib
,
410 Elf64_Sym
*sym
= find_symbol64(lib
, symname
);
413 printk(KERN_WARNING
"vDSO64: function %s not found !\n",
417 #ifdef VDS64_HAS_DESCRIPTORS
418 return *((u64
*)(vdso64_kbase
+ sym
->st_value
- VDSO64_LBASE
)) -
421 return sym
->st_value
- VDSO64_LBASE
;
425 static int vdso_do_func_patch64(struct lib32_elfinfo
*v32
,
426 struct lib64_elfinfo
*v64
,
427 const char *orig
, const char *fix
)
429 Elf64_Sym
*sym64_gen
, *sym64_fix
;
431 sym64_gen
= find_symbol64(v64
, orig
);
432 if (sym64_gen
== NULL
) {
433 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", orig
);
437 sym64_gen
->st_name
= 0;
440 sym64_fix
= find_symbol64(v64
, fix
);
441 if (sym64_fix
== NULL
) {
442 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", fix
);
445 sym64_gen
->st_value
= sym64_fix
->st_value
;
446 sym64_gen
->st_size
= sym64_fix
->st_size
;
447 sym64_gen
->st_info
= sym64_fix
->st_info
;
448 sym64_gen
->st_other
= sym64_fix
->st_other
;
449 sym64_gen
->st_shndx
= sym64_fix
->st_shndx
;
454 #endif /* CONFIG_PPC64 */
457 static __init
int vdso_do_find_sections(struct lib32_elfinfo
*v32
,
458 struct lib64_elfinfo
*v64
)
463 * Locate symbol tables & text section
466 v32
->dynsym
= find_section32(v32
->hdr
, ".dynsym", &v32
->dynsymsize
);
467 v32
->dynstr
= find_section32(v32
->hdr
, ".dynstr", NULL
);
468 if (v32
->dynsym
== NULL
|| v32
->dynstr
== NULL
) {
469 printk(KERN_ERR
"vDSO32: required symbol section not found\n");
472 sect
= find_section32(v32
->hdr
, ".text", NULL
);
474 printk(KERN_ERR
"vDSO32: the .text section was not found\n");
477 v32
->text
= sect
- vdso32_kbase
;
480 v64
->dynsym
= find_section64(v64
->hdr
, ".dynsym", &v64
->dynsymsize
);
481 v64
->dynstr
= find_section64(v64
->hdr
, ".dynstr", NULL
);
482 if (v64
->dynsym
== NULL
|| v64
->dynstr
== NULL
) {
483 printk(KERN_ERR
"vDSO64: required symbol section not found\n");
486 sect
= find_section64(v64
->hdr
, ".text", NULL
);
488 printk(KERN_ERR
"vDSO64: the .text section was not found\n");
491 v64
->text
= sect
- vdso64_kbase
;
492 #endif /* CONFIG_PPC64 */
497 static __init
void vdso_setup_trampolines(struct lib32_elfinfo
*v32
,
498 struct lib64_elfinfo
*v64
)
501 * Find signal trampolines
505 vdso64_rt_sigtramp
= find_function64(v64
, "__kernel_sigtramp_rt64");
507 vdso32_sigtramp
= find_function32(v32
, "__kernel_sigtramp32");
508 vdso32_rt_sigtramp
= find_function32(v32
, "__kernel_sigtramp_rt32");
511 static __init
int vdso_fixup_datapage(struct lib32_elfinfo
*v32
,
512 struct lib64_elfinfo
*v64
)
518 sym64
= find_symbol64(v64
, "__kernel_datapage_offset");
520 printk(KERN_ERR
"vDSO64: Can't find symbol "
521 "__kernel_datapage_offset !\n");
524 *((int *)(vdso64_kbase
+ sym64
->st_value
- VDSO64_LBASE
)) =
525 (vdso64_pages
<< PAGE_SHIFT
) -
526 (sym64
->st_value
- VDSO64_LBASE
);
527 #endif /* CONFIG_PPC64 */
529 sym32
= find_symbol32(v32
, "__kernel_datapage_offset");
531 printk(KERN_ERR
"vDSO32: Can't find symbol "
532 "__kernel_datapage_offset !\n");
535 *((int *)(vdso32_kbase
+ (sym32
->st_value
- VDSO32_LBASE
))) =
536 (vdso32_pages
<< PAGE_SHIFT
) -
537 (sym32
->st_value
- VDSO32_LBASE
);
543 static __init
int vdso_fixup_features(struct lib32_elfinfo
*v32
,
544 struct lib64_elfinfo
*v64
)
547 unsigned long size32
;
551 unsigned long size64
;
553 start64
= find_section64(v64
->hdr
, "__ftr_fixup", &size64
);
555 do_feature_fixups(cur_cpu_spec
->cpu_features
,
556 start64
, start64
+ size64
);
558 start64
= find_section64(v64
->hdr
, "__fw_ftr_fixup", &size64
);
560 do_feature_fixups(powerpc_firmware_features
,
561 start64
, start64
+ size64
);
562 #endif /* CONFIG_PPC64 */
564 start32
= find_section32(v32
->hdr
, "__ftr_fixup", &size32
);
566 do_feature_fixups(cur_cpu_spec
->cpu_features
,
567 start32
, start32
+ size32
);
570 start32
= find_section32(v32
->hdr
, "__fw_ftr_fixup", &size32
);
572 do_feature_fixups(powerpc_firmware_features
,
573 start32
, start32
+ size32
);
574 #endif /* CONFIG_PPC64 */
579 static __init
int vdso_fixup_alt_funcs(struct lib32_elfinfo
*v32
,
580 struct lib64_elfinfo
*v64
)
584 for (i
= 0; i
< ARRAY_SIZE(vdso_patches
); i
++) {
585 struct vdso_patch_def
*patch
= &vdso_patches
[i
];
586 int match
= (cur_cpu_spec
->cpu_features
& patch
->ftr_mask
)
591 DBG("replacing %s with %s...\n", patch
->gen_name
,
592 patch
->fix_name
? "NONE" : patch
->fix_name
);
595 * Patch the 32 bits and 64 bits symbols. Note that we do not
596 * patch the "." symbol on 64 bits.
597 * It would be easy to do, but doesn't seem to be necessary,
598 * patching the OPD symbol is enough.
600 vdso_do_func_patch32(v32
, v64
, patch
->gen_name
,
603 vdso_do_func_patch64(v32
, v64
, patch
->gen_name
,
605 #endif /* CONFIG_PPC64 */
612 static __init
int vdso_setup(void)
614 struct lib32_elfinfo v32
;
615 struct lib64_elfinfo v64
;
617 v32
.hdr
= vdso32_kbase
;
619 v64
.hdr
= vdso64_kbase
;
621 if (vdso_do_find_sections(&v32
, &v64
))
624 if (vdso_fixup_datapage(&v32
, &v64
))
627 if (vdso_fixup_features(&v32
, &v64
))
630 if (vdso_fixup_alt_funcs(&v32
, &v64
))
633 vdso_setup_trampolines(&v32
, &v64
);
639 * Called from setup_arch to initialize the bitmap of available
640 * syscalls in the systemcfg page
642 static void __init
vdso_setup_syscall_map(void)
645 extern unsigned long *sys_call_table
;
646 extern unsigned long sys_ni_syscall
;
649 for (i
= 0; i
< __NR_syscalls
; i
++) {
651 if (sys_call_table
[i
*2] != sys_ni_syscall
)
652 vdso_data
->syscall_map_64
[i
>> 5] |=
653 0x80000000UL
>> (i
& 0x1f);
654 if (sys_call_table
[i
*2+1] != sys_ni_syscall
)
655 vdso_data
->syscall_map_32
[i
>> 5] |=
656 0x80000000UL
>> (i
& 0x1f);
657 #else /* CONFIG_PPC64 */
658 if (sys_call_table
[i
] != sys_ni_syscall
)
659 vdso_data
->syscall_map_32
[i
>> 5] |=
660 0x80000000UL
>> (i
& 0x1f);
661 #endif /* CONFIG_PPC64 */
666 static int __init
vdso_init(void)
672 * Fill up the "systemcfg" stuff for backward compatiblity
674 strcpy(vdso_data
->eye_catcher
, "SYSTEMCFG:PPC64");
675 vdso_data
->version
.major
= SYSTEMCFG_MAJOR
;
676 vdso_data
->version
.minor
= SYSTEMCFG_MINOR
;
677 vdso_data
->processor
= mfspr(SPRN_PVR
);
679 * Fake the old platform number for pSeries and iSeries and add
680 * in LPAR bit if necessary
682 vdso_data
->platform
= machine_is(iseries
) ? 0x200 : 0x100;
683 if (firmware_has_feature(FW_FEATURE_LPAR
))
684 vdso_data
->platform
|= 1;
685 vdso_data
->physicalMemorySize
= lmb_phys_mem_size();
686 vdso_data
->dcache_size
= ppc64_caches
.dsize
;
687 vdso_data
->dcache_line_size
= ppc64_caches
.dline_size
;
688 vdso_data
->icache_size
= ppc64_caches
.isize
;
689 vdso_data
->icache_line_size
= ppc64_caches
.iline_size
;
692 * Calculate the size of the 64 bits vDSO
694 vdso64_pages
= (&vdso64_end
- &vdso64_start
) >> PAGE_SHIFT
;
695 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase
, vdso64_pages
);
696 #endif /* CONFIG_PPC64 */
700 * Calculate the size of the 32 bits vDSO
702 vdso32_pages
= (&vdso32_end
- &vdso32_start
) >> PAGE_SHIFT
;
703 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase
, vdso32_pages
);
707 * Setup the syscall map in the vDOS
709 vdso_setup_syscall_map();
712 * Initialize the vDSO images in memory, that is do necessary
713 * fixups of vDSO symbols, locate trampolines, etc...
716 printk(KERN_ERR
"vDSO setup failure, not enabled !\n");
724 /* Make sure pages are in the correct state */
725 vdso32_pagelist
= kzalloc(sizeof(struct page
*) * (vdso32_pages
+ 2),
727 BUG_ON(vdso32_pagelist
== NULL
);
728 for (i
= 0; i
< vdso32_pages
; i
++) {
729 struct page
*pg
= virt_to_page(vdso32_kbase
+ i
*PAGE_SIZE
);
730 ClearPageReserved(pg
);
732 vdso32_pagelist
[i
] = pg
;
734 vdso32_pagelist
[i
++] = virt_to_page(vdso_data
);
735 vdso32_pagelist
[i
] = NULL
;
738 vdso64_pagelist
= kzalloc(sizeof(struct page
*) * (vdso64_pages
+ 2),
740 BUG_ON(vdso64_pagelist
== NULL
);
741 for (i
= 0; i
< vdso64_pages
; i
++) {
742 struct page
*pg
= virt_to_page(vdso64_kbase
+ i
*PAGE_SIZE
);
743 ClearPageReserved(pg
);
745 vdso64_pagelist
[i
] = pg
;
747 vdso64_pagelist
[i
++] = virt_to_page(vdso_data
);
748 vdso64_pagelist
[i
] = NULL
;
749 #endif /* CONFIG_PPC64 */
751 get_page(virt_to_page(vdso_data
));
758 arch_initcall(vdso_init
);
760 int in_gate_area_no_task(unsigned long addr
)
765 int in_gate_area(struct task_struct
*task
, unsigned long addr
)
770 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)