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>
34 #include <asm/machdep.h>
35 #include <asm/cputable.h>
36 #include <asm/sections.h>
42 #define DBG(fmt...) printk(fmt)
49 * The vDSOs themselves are here
51 extern char vdso64_start
, vdso64_end
;
52 extern char vdso32_start
, vdso32_end
;
54 static void *vdso64_kbase
= &vdso64_start
;
55 static void *vdso32_kbase
= &vdso32_start
;
57 unsigned int vdso64_pages
;
58 unsigned int vdso32_pages
;
60 /* Signal trampolines user addresses */
62 unsigned long vdso64_rt_sigtramp
;
63 unsigned long vdso32_sigtramp
;
64 unsigned long vdso32_rt_sigtramp
;
66 /* Format of the patch table */
69 u32 pvr_mask
, pvr_value
;
74 /* Table of functions to patch based on the CPU type/revision
76 * TODO: Improve by adding whole lists for each entry
78 static struct vdso_patch_def vdso_patches
[] = {
80 0xffff0000, 0x003a0000, /* POWER5 */
81 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
84 0xffff0000, 0x003b0000, /* POWER5 */
85 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
90 * Some infos carried around for each of them during parsing at
95 Elf32_Ehdr
*hdr
; /* ptr to ELF */
96 Elf32_Sym
*dynsym
; /* ptr to .dynsym section */
97 unsigned long dynsymsize
; /* size of .dynsym section */
98 char *dynstr
; /* ptr to .dynstr section */
99 unsigned long text
; /* offset of .text section in .so */
106 unsigned long dynsymsize
;
113 static void dump_one_vdso_page(struct page
*pg
, struct page
*upg
)
115 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg
) << PAGE_SHIFT
),
118 if (upg
/* && pg != upg*/) {
119 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg
) << PAGE_SHIFT
),
126 static void dump_vdso_pages(struct vm_area_struct
* vma
)
130 if (!vma
|| test_thread_flag(TIF_32BIT
)) {
131 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase
);
132 for (i
=0; i
<vdso32_pages
; i
++) {
133 struct page
*pg
= virt_to_page(vdso32_kbase
+ i
*PAGE_SIZE
);
134 struct page
*upg
= (vma
&& vma
->vm_mm
) ?
135 follow_page(vma
->vm_mm
, vma
->vm_start
+ i
*PAGE_SIZE
, 0)
137 dump_one_vdso_page(pg
, upg
);
140 if (!vma
|| !test_thread_flag(TIF_32BIT
)) {
141 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase
);
142 for (i
=0; i
<vdso64_pages
; i
++) {
143 struct page
*pg
= virt_to_page(vdso64_kbase
+ i
*PAGE_SIZE
);
144 struct page
*upg
= (vma
&& vma
->vm_mm
) ?
145 follow_page(vma
->vm_mm
, vma
->vm_start
+ i
*PAGE_SIZE
, 0)
147 dump_one_vdso_page(pg
, upg
);
154 * Keep a dummy vma_close for now, it will prevent VMA merging.
156 static void vdso_vma_close(struct vm_area_struct
* vma
)
161 * Our nopage() function, maps in the actual vDSO kernel pages, they will
162 * be mapped read-only by do_no_page(), and eventually COW'ed, either
163 * right away for an initial write access, or by do_wp_page().
165 static struct page
* vdso_vma_nopage(struct vm_area_struct
* vma
,
166 unsigned long address
, int *type
)
168 unsigned long offset
= address
- vma
->vm_start
;
170 void *vbase
= test_thread_flag(TIF_32BIT
) ? vdso32_kbase
: vdso64_kbase
;
172 DBG("vdso_vma_nopage(current: %s, address: %016lx, off: %lx)\n",
173 current
->comm
, address
, offset
);
175 if (address
< vma
->vm_start
|| address
> vma
->vm_end
)
176 return NOPAGE_SIGBUS
;
179 * Last page is systemcfg, special handling here, no get_page() a
180 * this is a reserved page
182 if ((vma
->vm_end
- address
) <= PAGE_SIZE
)
183 return virt_to_page(systemcfg
);
185 pg
= virt_to_page(vbase
+ offset
);
187 DBG(" ->page count: %d\n", page_count(pg
));
192 static struct vm_operations_struct vdso_vmops
= {
193 .close
= vdso_vma_close
,
194 .nopage
= vdso_vma_nopage
,
198 * This is called from binfmt_elf, we create the special vma for the
199 * vDSO and insert it into the mm struct tree
201 int arch_setup_additional_pages(struct linux_binprm
*bprm
, int executable_stack
)
203 struct mm_struct
*mm
= current
->mm
;
204 struct vm_area_struct
*vma
;
205 unsigned long vdso_pages
;
206 unsigned long vdso_base
;
208 if (test_thread_flag(TIF_32BIT
)) {
209 vdso_pages
= vdso32_pages
;
210 vdso_base
= VDSO32_MBASE
;
212 vdso_pages
= vdso64_pages
;
213 vdso_base
= VDSO64_MBASE
;
216 /* vDSO has a problem and was disabled, just don't "enable" it for the
219 if (vdso_pages
== 0) {
220 current
->thread
.vdso_base
= 0;
223 vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
226 if (security_vm_enough_memory(vdso_pages
)) {
227 kmem_cache_free(vm_area_cachep
, vma
);
230 memset(vma
, 0, sizeof(*vma
));
233 * pick a base address for the vDSO in process space. We have a default
234 * base of 1Mb on which we had a random offset up to 1Mb.
235 * XXX: Add possibility for a program header to specify that location
237 current
->thread
.vdso_base
= vdso_base
;
238 /* + ((unsigned long)vma & 0x000ff000); */
241 vma
->vm_start
= current
->thread
.vdso_base
;
244 * the VMA size is one page more than the vDSO since systemcfg
245 * is mapped in the last one
247 vma
->vm_end
= vma
->vm_start
+ ((vdso_pages
+ 1) << PAGE_SHIFT
);
250 * our vma flags don't have VM_WRITE so by default, the process isn't allowed
251 * to write those pages.
252 * gdb can break that with ptrace interface, and thus trigger COW on those
253 * pages but it's then your responsibility to never do that on the "data" page
254 * of the vDSO or you'll stop getting kernel updates and your nice userland
255 * gettimeofday will be totally dead. It's fine to use that for setting
256 * breakpoints in the vDSO code pages though
258 vma
->vm_flags
= VM_READ
| VM_EXEC
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
259 vma
->vm_flags
|= mm
->def_flags
;
260 vma
->vm_page_prot
= protection_map
[vma
->vm_flags
& 0x7];
261 vma
->vm_ops
= &vdso_vmops
;
263 down_write(&mm
->mmap_sem
);
264 insert_vm_struct(mm
, vma
);
265 mm
->total_vm
+= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
266 up_write(&mm
->mmap_sem
);
271 static void * __init
find_section32(Elf32_Ehdr
*ehdr
, const char *secname
,
278 /* Grab section headers and strings so we can tell who is who */
279 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
280 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
282 /* Find the section they want */
283 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
284 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
286 *size
= sechdrs
[i
].sh_size
;
287 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
294 static void * __init
find_section64(Elf64_Ehdr
*ehdr
, const char *secname
,
301 /* Grab section headers and strings so we can tell who is who */
302 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
303 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
305 /* Find the section they want */
306 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
307 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
309 *size
= sechdrs
[i
].sh_size
;
310 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
318 static Elf32_Sym
* __init
find_symbol32(struct lib32_elfinfo
*lib
, const char *symname
)
323 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf32_Sym
)); i
++) {
324 if (lib
->dynsym
[i
].st_name
== 0)
326 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
, 32);
327 c
= strchr(name
, '@');
330 if (strcmp(symname
, name
) == 0)
331 return &lib
->dynsym
[i
];
336 static Elf64_Sym
* __init
find_symbol64(struct lib64_elfinfo
*lib
, const char *symname
)
341 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf64_Sym
)); i
++) {
342 if (lib
->dynsym
[i
].st_name
== 0)
344 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
, 32);
345 c
= strchr(name
, '@');
348 if (strcmp(symname
, name
) == 0)
349 return &lib
->dynsym
[i
];
354 /* Note that we assume the section is .text and the symbol is relative to
357 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
, const char *symname
)
359 Elf32_Sym
*sym
= find_symbol32(lib
, symname
);
362 printk(KERN_WARNING
"vDSO32: function %s not found !\n", symname
);
365 return sym
->st_value
- VDSO32_LBASE
;
368 /* Note that we assume the section is .text and the symbol is relative to
371 static unsigned long __init
find_function64(struct lib64_elfinfo
*lib
, const char *symname
)
373 Elf64_Sym
*sym
= find_symbol64(lib
, symname
);
376 printk(KERN_WARNING
"vDSO64: function %s not found !\n", symname
);
379 #ifdef VDS64_HAS_DESCRIPTORS
380 return *((u64
*)(vdso64_kbase
+ sym
->st_value
- VDSO64_LBASE
)) - VDSO64_LBASE
;
382 return sym
->st_value
- VDSO64_LBASE
;
387 static __init
int vdso_do_find_sections(struct lib32_elfinfo
*v32
,
388 struct lib64_elfinfo
*v64
)
393 * Locate symbol tables & text section
396 v32
->dynsym
= find_section32(v32
->hdr
, ".dynsym", &v32
->dynsymsize
);
397 v32
->dynstr
= find_section32(v32
->hdr
, ".dynstr", NULL
);
398 if (v32
->dynsym
== NULL
|| v32
->dynstr
== NULL
) {
399 printk(KERN_ERR
"vDSO32: a required symbol section was not found\n");
402 sect
= find_section32(v32
->hdr
, ".text", NULL
);
404 printk(KERN_ERR
"vDSO32: the .text section was not found\n");
407 v32
->text
= sect
- vdso32_kbase
;
409 v64
->dynsym
= find_section64(v64
->hdr
, ".dynsym", &v64
->dynsymsize
);
410 v64
->dynstr
= find_section64(v64
->hdr
, ".dynstr", NULL
);
411 if (v64
->dynsym
== NULL
|| v64
->dynstr
== NULL
) {
412 printk(KERN_ERR
"vDSO64: a required symbol section was not found\n");
415 sect
= find_section64(v64
->hdr
, ".text", NULL
);
417 printk(KERN_ERR
"vDSO64: the .text section was not found\n");
420 v64
->text
= sect
- vdso64_kbase
;
425 static __init
void vdso_setup_trampolines(struct lib32_elfinfo
*v32
,
426 struct lib64_elfinfo
*v64
)
429 * Find signal trampolines
432 vdso64_rt_sigtramp
= find_function64(v64
, "__kernel_sigtramp_rt64");
433 vdso32_sigtramp
= find_function32(v32
, "__kernel_sigtramp32");
434 vdso32_rt_sigtramp
= find_function32(v32
, "__kernel_sigtramp_rt32");
437 static __init
int vdso_fixup_datapage(struct lib32_elfinfo
*v32
,
438 struct lib64_elfinfo
*v64
)
443 sym32
= find_symbol32(v32
, "__kernel_datapage_offset");
445 printk(KERN_ERR
"vDSO32: Can't find symbol __kernel_datapage_offset !\n");
448 *((int *)(vdso32_kbase
+ (sym32
->st_value
- VDSO32_LBASE
))) =
449 (vdso32_pages
<< PAGE_SHIFT
) - (sym32
->st_value
- VDSO32_LBASE
);
451 sym64
= find_symbol64(v64
, "__kernel_datapage_offset");
453 printk(KERN_ERR
"vDSO64: Can't find symbol __kernel_datapage_offset !\n");
456 *((int *)(vdso64_kbase
+ sym64
->st_value
- VDSO64_LBASE
)) =
457 (vdso64_pages
<< PAGE_SHIFT
) - (sym64
->st_value
- VDSO64_LBASE
);
462 static int vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
463 struct lib64_elfinfo
*v64
,
464 const char *orig
, const char *fix
)
466 Elf32_Sym
*sym32_gen
, *sym32_fix
;
468 sym32_gen
= find_symbol32(v32
, orig
);
469 if (sym32_gen
== NULL
) {
470 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", orig
);
473 sym32_fix
= find_symbol32(v32
, fix
);
474 if (sym32_fix
== NULL
) {
475 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", fix
);
478 sym32_gen
->st_value
= sym32_fix
->st_value
;
479 sym32_gen
->st_size
= sym32_fix
->st_size
;
480 sym32_gen
->st_info
= sym32_fix
->st_info
;
481 sym32_gen
->st_other
= sym32_fix
->st_other
;
482 sym32_gen
->st_shndx
= sym32_fix
->st_shndx
;
487 static int vdso_do_func_patch64(struct lib32_elfinfo
*v32
,
488 struct lib64_elfinfo
*v64
,
489 const char *orig
, const char *fix
)
491 Elf64_Sym
*sym64_gen
, *sym64_fix
;
493 sym64_gen
= find_symbol64(v64
, orig
);
494 if (sym64_gen
== NULL
) {
495 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", orig
);
498 sym64_fix
= find_symbol64(v64
, fix
);
499 if (sym64_fix
== NULL
) {
500 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", fix
);
503 sym64_gen
->st_value
= sym64_fix
->st_value
;
504 sym64_gen
->st_size
= sym64_fix
->st_size
;
505 sym64_gen
->st_info
= sym64_fix
->st_info
;
506 sym64_gen
->st_other
= sym64_fix
->st_other
;
507 sym64_gen
->st_shndx
= sym64_fix
->st_shndx
;
512 static __init
int vdso_fixup_alt_funcs(struct lib32_elfinfo
*v32
,
513 struct lib64_elfinfo
*v64
)
518 pvr
= mfspr(SPRN_PVR
);
519 for (i
= 0; i
< ARRAY_SIZE(vdso_patches
); i
++) {
520 struct vdso_patch_def
*patch
= &vdso_patches
[i
];
521 int match
= (pvr
& patch
->pvr_mask
) == patch
->pvr_value
;
523 DBG("patch %d (mask: %x, pvr: %x) : %s\n",
524 i
, patch
->pvr_mask
, patch
->pvr_value
, match
? "match" : "skip");
529 DBG("replacing %s with %s...\n", patch
->gen_name
, patch
->fix_name
);
532 * Patch the 32 bits and 64 bits symbols. Note that we do not patch
533 * the "." symbol on 64 bits. It would be easy to do, but doesn't
534 * seem to be necessary, patching the OPD symbol is enough.
536 vdso_do_func_patch32(v32
, v64
, patch
->gen_name
, patch
->fix_name
);
537 vdso_do_func_patch64(v32
, v64
, patch
->gen_name
, patch
->fix_name
);
544 static __init
int vdso_setup(void)
546 struct lib32_elfinfo v32
;
547 struct lib64_elfinfo v64
;
549 v32
.hdr
= vdso32_kbase
;
550 v64
.hdr
= vdso64_kbase
;
552 if (vdso_do_find_sections(&v32
, &v64
))
555 if (vdso_fixup_datapage(&v32
, &v64
))
558 if (vdso_fixup_alt_funcs(&v32
, &v64
))
561 vdso_setup_trampolines(&v32
, &v64
);
566 void __init
vdso_init(void)
570 vdso64_pages
= (&vdso64_end
- &vdso64_start
) >> PAGE_SHIFT
;
571 vdso32_pages
= (&vdso32_end
- &vdso32_start
) >> PAGE_SHIFT
;
573 DBG("vdso64_kbase: %p, 0x%x pages, vdso32_kbase: %p, 0x%x pages\n",
574 vdso64_kbase
, vdso64_pages
, vdso32_kbase
, vdso32_pages
);
577 * Initialize the vDSO images in memory, that is do necessary
578 * fixups of vDSO symbols, locate trampolines, etc...
581 printk(KERN_ERR
"vDSO setup failure, not enabled !\n");
582 /* XXX should free pages here ? */
583 vdso64_pages
= vdso32_pages
= 0;
587 /* Make sure pages are in the correct state */
588 for (i
= 0; i
< vdso64_pages
; i
++) {
589 struct page
*pg
= virt_to_page(vdso64_kbase
+ i
*PAGE_SIZE
);
590 ClearPageReserved(pg
);
593 for (i
= 0; i
< vdso32_pages
; i
++) {
594 struct page
*pg
= virt_to_page(vdso32_kbase
+ i
*PAGE_SIZE
);
595 ClearPageReserved(pg
);
600 int in_gate_area_no_task(unsigned long addr
)
605 int in_gate_area(struct task_struct
*task
, unsigned long addr
)
610 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)