2 * linux/arch/ppc64/kernel/vdso.c
4 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
5 * <benh@kernel.crashing.org>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
13 #include <linux/config.h>
14 #include <linux/module.h>
15 #include <linux/errno.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
19 #include <linux/smp.h>
20 #include <linux/smp_lock.h>
21 #include <linux/stddef.h>
22 #include <linux/unistd.h>
23 #include <linux/slab.h>
24 #include <linux/user.h>
25 #include <linux/elf.h>
26 #include <linux/security.h>
27 #include <linux/bootmem.h>
29 #include <asm/pgtable.h>
30 #include <asm/system.h>
31 #include <asm/processor.h>
33 #include <asm/mmu_context.h>
35 #include <asm/machdep.h>
36 #include <asm/cputable.h>
37 #include <asm/sections.h>
39 #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 unsigned int vdso32_pages
;
55 unsigned long vdso32_sigtramp
;
56 unsigned long vdso32_rt_sigtramp
;
59 extern char vdso64_start
, vdso64_end
;
60 static void *vdso64_kbase
= &vdso64_start
;
61 unsigned int vdso64_pages
;
62 unsigned long vdso64_rt_sigtramp
;
63 #endif /* CONFIG_PPC64 */
66 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
67 * Once the early boot kernel code no longer needs to muck around
68 * with it, it will become dynamically allocated
71 struct vdso_data data
;
73 } vdso_data_store
__attribute__((__section__(".data.page_aligned")));
74 struct vdso_data
*vdso_data
= &vdso_data_store
.data
;
76 /* Format of the patch table */
79 unsigned long ftr_mask
, ftr_value
;
84 /* Table of functions to patch based on the CPU type/revision
86 * Currently, we only change sync_dicache to do nothing on processors
87 * with a coherent icache
89 static struct vdso_patch_def vdso_patches
[] = {
91 CPU_FTR_COHERENT_ICACHE
, CPU_FTR_COHERENT_ICACHE
,
92 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
96 "__kernel_gettimeofday", NULL
101 * Some infos carried around for each of them during parsing at
106 Elf32_Ehdr
*hdr
; /* ptr to ELF */
107 Elf32_Sym
*dynsym
; /* ptr to .dynsym section */
108 unsigned long dynsymsize
; /* size of .dynsym section */
109 char *dynstr
; /* ptr to .dynstr section */
110 unsigned long text
; /* offset of .text section in .so */
117 unsigned long dynsymsize
;
124 static void dump_one_vdso_page(struct page
*pg
, struct page
*upg
)
126 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg
) << PAGE_SHIFT
),
129 if (upg
/* && pg != upg*/) {
130 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg
)
138 static void dump_vdso_pages(struct vm_area_struct
* vma
)
142 if (!vma
|| test_thread_flag(TIF_32BIT
)) {
143 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase
);
144 for (i
=0; i
<vdso32_pages
; i
++) {
145 struct page
*pg
= virt_to_page(vdso32_kbase
+
147 struct page
*upg
= (vma
&& vma
->vm_mm
) ?
148 follow_page(vma
->vm_mm
, vma
->vm_start
+
151 dump_one_vdso_page(pg
, upg
);
154 if (!vma
|| !test_thread_flag(TIF_32BIT
)) {
155 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase
);
156 for (i
=0; i
<vdso64_pages
; i
++) {
157 struct page
*pg
= virt_to_page(vdso64_kbase
+
159 struct page
*upg
= (vma
&& vma
->vm_mm
) ?
160 follow_page(vma
->vm_mm
, vma
->vm_start
+
163 dump_one_vdso_page(pg
, upg
);
170 * Keep a dummy vma_close for now, it will prevent VMA merging.
172 static void vdso_vma_close(struct vm_area_struct
* vma
)
177 * Our nopage() function, maps in the actual vDSO kernel pages, they will
178 * be mapped read-only by do_no_page(), and eventually COW'ed, either
179 * right away for an initial write access, or by do_wp_page().
181 static struct page
* vdso_vma_nopage(struct vm_area_struct
* vma
,
182 unsigned long address
, int *type
)
184 unsigned long offset
= address
- vma
->vm_start
;
187 void *vbase
= test_thread_flag(TIF_32BIT
) ?
188 vdso32_kbase
: vdso64_kbase
;
190 void *vbase
= vdso32_kbase
;
193 DBG("vdso_vma_nopage(current: %s, address: %016lx, off: %lx)\n",
194 current
->comm
, address
, offset
);
196 if (address
< vma
->vm_start
|| address
> vma
->vm_end
)
197 return NOPAGE_SIGBUS
;
200 * Last page is systemcfg.
202 if ((vma
->vm_end
- address
) <= PAGE_SIZE
)
203 pg
= virt_to_page(vdso_data
);
205 pg
= virt_to_page(vbase
+ offset
);
208 DBG(" ->page count: %d\n", page_count(pg
));
213 static struct vm_operations_struct vdso_vmops
= {
214 .close
= vdso_vma_close
,
215 .nopage
= vdso_vma_nopage
,
219 * This is called from binfmt_elf, we create the special vma for the
220 * vDSO and insert it into the mm struct tree
222 int arch_setup_additional_pages(struct linux_binprm
*bprm
,
223 int executable_stack
)
225 struct mm_struct
*mm
= current
->mm
;
226 struct vm_area_struct
*vma
;
227 unsigned long vdso_pages
;
228 unsigned long vdso_base
;
231 if (test_thread_flag(TIF_32BIT
)) {
232 vdso_pages
= vdso32_pages
;
233 vdso_base
= VDSO32_MBASE
;
235 vdso_pages
= vdso64_pages
;
236 vdso_base
= VDSO64_MBASE
;
239 vdso_pages
= vdso32_pages
;
240 vdso_base
= VDSO32_MBASE
;
243 current
->thread
.vdso_base
= 0;
245 /* vDSO has a problem and was disabled, just don't "enable" it for the
251 vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
255 memset(vma
, 0, sizeof(*vma
));
257 /* Add a page to the vdso size for the data page */
261 * pick a base address for the vDSO in process space. We try to put it
262 * at vdso_base which is the "natural" base for it, but we might fail
263 * and end up putting it elsewhere.
265 vdso_base
= get_unmapped_area(NULL
, vdso_base
,
266 vdso_pages
<< PAGE_SHIFT
, 0, 0);
267 if (vdso_base
& ~PAGE_MASK
) {
268 kmem_cache_free(vm_area_cachep
, vma
);
269 return (int)vdso_base
;
272 current
->thread
.vdso_base
= vdso_base
;
275 vma
->vm_start
= current
->thread
.vdso_base
;
276 vma
->vm_end
= vma
->vm_start
+ (vdso_pages
<< PAGE_SHIFT
);
279 * our vma flags don't have VM_WRITE so by default, the process isn't
280 * allowed to write those pages.
281 * gdb can break that with ptrace interface, and thus trigger COW on
282 * those pages but it's then your responsibility to never do that on
283 * the "data" page of the vDSO or you'll stop getting kernel updates
284 * and your nice userland gettimeofday will be totally dead.
285 * It's fine to use that for setting breakpoints in the vDSO code
288 vma
->vm_flags
= VM_READ
| VM_EXEC
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
289 vma
->vm_flags
|= mm
->def_flags
;
290 vma
->vm_page_prot
= protection_map
[vma
->vm_flags
& 0x7];
291 vma
->vm_ops
= &vdso_vmops
;
293 down_write(&mm
->mmap_sem
);
294 if (insert_vm_struct(mm
, vma
)) {
295 up_write(&mm
->mmap_sem
);
296 kmem_cache_free(vm_area_cachep
, vma
);
299 mm
->total_vm
+= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
300 up_write(&mm
->mmap_sem
);
305 static void * __init
find_section32(Elf32_Ehdr
*ehdr
, const char *secname
,
312 /* Grab section headers and strings so we can tell who is who */
313 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
314 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
316 /* Find the section they want */
317 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
318 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
320 *size
= sechdrs
[i
].sh_size
;
321 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
328 static Elf32_Sym
* __init
find_symbol32(struct lib32_elfinfo
*lib
,
332 char name
[MAX_SYMNAME
], *c
;
334 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf32_Sym
)); i
++) {
335 if (lib
->dynsym
[i
].st_name
== 0)
337 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
339 c
= strchr(name
, '@');
342 if (strcmp(symname
, name
) == 0)
343 return &lib
->dynsym
[i
];
348 /* Note that we assume the section is .text and the symbol is relative to
351 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
354 Elf32_Sym
*sym
= find_symbol32(lib
, symname
);
357 printk(KERN_WARNING
"vDSO32: function %s not found !\n",
361 return sym
->st_value
- VDSO32_LBASE
;
364 static int vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
365 struct lib64_elfinfo
*v64
,
366 const char *orig
, const char *fix
)
368 Elf32_Sym
*sym32_gen
, *sym32_fix
;
370 sym32_gen
= find_symbol32(v32
, orig
);
371 if (sym32_gen
== NULL
) {
372 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", orig
);
376 sym32_gen
->st_name
= 0;
379 sym32_fix
= find_symbol32(v32
, fix
);
380 if (sym32_fix
== NULL
) {
381 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", fix
);
384 sym32_gen
->st_value
= sym32_fix
->st_value
;
385 sym32_gen
->st_size
= sym32_fix
->st_size
;
386 sym32_gen
->st_info
= sym32_fix
->st_info
;
387 sym32_gen
->st_other
= sym32_fix
->st_other
;
388 sym32_gen
->st_shndx
= sym32_fix
->st_shndx
;
396 static void * __init
find_section64(Elf64_Ehdr
*ehdr
, const char *secname
,
403 /* Grab section headers and strings so we can tell who is who */
404 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
405 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
407 /* Find the section they want */
408 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
409 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
411 *size
= sechdrs
[i
].sh_size
;
412 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
420 static Elf64_Sym
* __init
find_symbol64(struct lib64_elfinfo
*lib
,
424 char name
[MAX_SYMNAME
], *c
;
426 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf64_Sym
)); i
++) {
427 if (lib
->dynsym
[i
].st_name
== 0)
429 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
431 c
= strchr(name
, '@');
434 if (strcmp(symname
, name
) == 0)
435 return &lib
->dynsym
[i
];
440 /* Note that we assume the section is .text and the symbol is relative to
443 static unsigned long __init
find_function64(struct lib64_elfinfo
*lib
,
446 Elf64_Sym
*sym
= find_symbol64(lib
, symname
);
449 printk(KERN_WARNING
"vDSO64: function %s not found !\n",
453 #ifdef VDS64_HAS_DESCRIPTORS
454 return *((u64
*)(vdso64_kbase
+ sym
->st_value
- VDSO64_LBASE
)) -
457 return sym
->st_value
- VDSO64_LBASE
;
461 static int vdso_do_func_patch64(struct lib32_elfinfo
*v32
,
462 struct lib64_elfinfo
*v64
,
463 const char *orig
, const char *fix
)
465 Elf64_Sym
*sym64_gen
, *sym64_fix
;
467 sym64_gen
= find_symbol64(v64
, orig
);
468 if (sym64_gen
== NULL
) {
469 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", orig
);
473 sym64_gen
->st_name
= 0;
476 sym64_fix
= find_symbol64(v64
, fix
);
477 if (sym64_fix
== NULL
) {
478 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", fix
);
481 sym64_gen
->st_value
= sym64_fix
->st_value
;
482 sym64_gen
->st_size
= sym64_fix
->st_size
;
483 sym64_gen
->st_info
= sym64_fix
->st_info
;
484 sym64_gen
->st_other
= sym64_fix
->st_other
;
485 sym64_gen
->st_shndx
= sym64_fix
->st_shndx
;
490 #endif /* CONFIG_PPC64 */
493 static __init
int vdso_do_find_sections(struct lib32_elfinfo
*v32
,
494 struct lib64_elfinfo
*v64
)
499 * Locate symbol tables & text section
502 v32
->dynsym
= find_section32(v32
->hdr
, ".dynsym", &v32
->dynsymsize
);
503 v32
->dynstr
= find_section32(v32
->hdr
, ".dynstr", NULL
);
504 if (v32
->dynsym
== NULL
|| v32
->dynstr
== NULL
) {
505 printk(KERN_ERR
"vDSO32: required symbol section not found\n");
508 sect
= find_section32(v32
->hdr
, ".text", NULL
);
510 printk(KERN_ERR
"vDSO32: the .text section was not found\n");
513 v32
->text
= sect
- vdso32_kbase
;
516 v64
->dynsym
= find_section64(v64
->hdr
, ".dynsym", &v64
->dynsymsize
);
517 v64
->dynstr
= find_section64(v64
->hdr
, ".dynstr", NULL
);
518 if (v64
->dynsym
== NULL
|| v64
->dynstr
== NULL
) {
519 printk(KERN_ERR
"vDSO64: required symbol section not found\n");
522 sect
= find_section64(v64
->hdr
, ".text", NULL
);
524 printk(KERN_ERR
"vDSO64: the .text section was not found\n");
527 v64
->text
= sect
- vdso64_kbase
;
528 #endif /* CONFIG_PPC64 */
533 static __init
void vdso_setup_trampolines(struct lib32_elfinfo
*v32
,
534 struct lib64_elfinfo
*v64
)
537 * Find signal trampolines
541 vdso64_rt_sigtramp
= find_function64(v64
, "__kernel_sigtramp_rt64");
543 vdso32_sigtramp
= find_function32(v32
, "__kernel_sigtramp32");
544 vdso32_rt_sigtramp
= find_function32(v32
, "__kernel_sigtramp_rt32");
547 static __init
int vdso_fixup_datapage(struct lib32_elfinfo
*v32
,
548 struct lib64_elfinfo
*v64
)
554 sym64
= find_symbol64(v64
, "__kernel_datapage_offset");
556 printk(KERN_ERR
"vDSO64: Can't find symbol "
557 "__kernel_datapage_offset !\n");
560 *((int *)(vdso64_kbase
+ sym64
->st_value
- VDSO64_LBASE
)) =
561 (vdso64_pages
<< PAGE_SHIFT
) -
562 (sym64
->st_value
- VDSO64_LBASE
);
563 #endif /* CONFIG_PPC64 */
565 sym32
= find_symbol32(v32
, "__kernel_datapage_offset");
567 printk(KERN_ERR
"vDSO32: Can't find symbol "
568 "__kernel_datapage_offset !\n");
571 *((int *)(vdso32_kbase
+ (sym32
->st_value
- VDSO32_LBASE
))) =
572 (vdso32_pages
<< PAGE_SHIFT
) -
573 (sym32
->st_value
- VDSO32_LBASE
);
578 static __init
int vdso_fixup_alt_funcs(struct lib32_elfinfo
*v32
,
579 struct lib64_elfinfo
*v64
)
583 for (i
= 0; i
< ARRAY_SIZE(vdso_patches
); i
++) {
584 struct vdso_patch_def
*patch
= &vdso_patches
[i
];
585 int match
= (cur_cpu_spec
->cpu_features
& patch
->ftr_mask
)
590 DBG("replacing %s with %s...\n", patch
->gen_name
,
591 patch
->fix_name
? "NONE" : patch
->fix_name
);
594 * Patch the 32 bits and 64 bits symbols. Note that we do not
595 * patch the "." symbol on 64 bits.
596 * It would be easy to do, but doesn't seem to be necessary,
597 * patching the OPD symbol is enough.
599 vdso_do_func_patch32(v32
, v64
, patch
->gen_name
,
602 vdso_do_func_patch64(v32
, v64
, patch
->gen_name
,
604 #endif /* CONFIG_PPC64 */
611 static __init
int vdso_setup(void)
613 struct lib32_elfinfo v32
;
614 struct lib64_elfinfo v64
;
616 v32
.hdr
= vdso32_kbase
;
618 v64
.hdr
= vdso64_kbase
;
620 if (vdso_do_find_sections(&v32
, &v64
))
623 if (vdso_fixup_datapage(&v32
, &v64
))
626 if (vdso_fixup_alt_funcs(&v32
, &v64
))
629 vdso_setup_trampolines(&v32
, &v64
);
635 * Called from setup_arch to initialize the bitmap of available
636 * syscalls in the systemcfg page
638 static void __init
vdso_setup_syscall_map(void)
641 extern unsigned long *sys_call_table
;
642 extern unsigned long sys_ni_syscall
;
645 for (i
= 0; i
< __NR_syscalls
; i
++) {
647 if (sys_call_table
[i
*2] != sys_ni_syscall
)
648 vdso_data
->syscall_map_64
[i
>> 5] |=
649 0x80000000UL
>> (i
& 0x1f);
650 if (sys_call_table
[i
*2+1] != sys_ni_syscall
)
651 vdso_data
->syscall_map_32
[i
>> 5] |=
652 0x80000000UL
>> (i
& 0x1f);
653 #else /* CONFIG_PPC64 */
654 if (sys_call_table
[i
] != sys_ni_syscall
)
655 vdso_data
->syscall_map_32
[i
>> 5] |=
656 0x80000000UL
>> (i
& 0x1f);
657 #endif /* CONFIG_PPC64 */
662 void __init
vdso_init(void)
668 * Fill up the "systemcfg" stuff for backward compatiblity
670 strcpy(vdso_data
->eye_catcher
, "SYSTEMCFG:PPC64");
671 vdso_data
->version
.major
= SYSTEMCFG_MAJOR
;
672 vdso_data
->version
.minor
= SYSTEMCFG_MINOR
;
673 vdso_data
->processor
= mfspr(SPRN_PVR
);
674 vdso_data
->platform
= _machine
;
675 vdso_data
->physicalMemorySize
= lmb_phys_mem_size();
676 vdso_data
->dcache_size
= ppc64_caches
.dsize
;
677 vdso_data
->dcache_line_size
= ppc64_caches
.dline_size
;
678 vdso_data
->icache_size
= ppc64_caches
.isize
;
679 vdso_data
->icache_line_size
= ppc64_caches
.iline_size
;
682 * Calculate the size of the 64 bits vDSO
684 vdso64_pages
= (&vdso64_end
- &vdso64_start
) >> PAGE_SHIFT
;
685 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase
, vdso64_pages
);
686 #endif /* CONFIG_PPC64 */
690 * Calculate the size of the 32 bits vDSO
692 vdso32_pages
= (&vdso32_end
- &vdso32_start
) >> PAGE_SHIFT
;
693 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase
, vdso32_pages
);
697 * Setup the syscall map in the vDOS
699 vdso_setup_syscall_map();
701 * Initialize the vDSO images in memory, that is do necessary
702 * fixups of vDSO symbols, locate trampolines, etc...
705 printk(KERN_ERR
"vDSO setup failure, not enabled !\n");
713 /* Make sure pages are in the correct state */
714 for (i
= 0; i
< vdso32_pages
; i
++) {
715 struct page
*pg
= virt_to_page(vdso32_kbase
+ i
*PAGE_SIZE
);
716 ClearPageReserved(pg
);
721 for (i
= 0; i
< vdso64_pages
; i
++) {
722 struct page
*pg
= virt_to_page(vdso64_kbase
+ i
*PAGE_SIZE
);
723 ClearPageReserved(pg
);
726 #endif /* CONFIG_PPC64 */
728 get_page(virt_to_page(vdso_data
));
731 int in_gate_area_no_task(unsigned long addr
)
736 int in_gate_area(struct task_struct
*task
, unsigned long addr
)
741 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)